WO2018195891A1 - Method and apparatus for evaluating quality of non-reference image - Google Patents

Method and apparatus for evaluating quality of non-reference image Download PDF

Info

Publication number
WO2018195891A1
WO2018195891A1 PCT/CN2017/082348 CN2017082348W WO2018195891A1 WO 2018195891 A1 WO2018195891 A1 WO 2018195891A1 CN 2017082348 W CN2017082348 W CN 2017082348W WO 2018195891 A1 WO2018195891 A1 WO 2018195891A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
quality score
distortion
region
distortion type
Prior art date
Application number
PCT/CN2017/082348
Other languages
French (fr)
Chinese (zh)
Inventor
朱映映
曹磊
王旭
江健民
Original Assignee
深圳大学
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 深圳大学 filed Critical 深圳大学
Priority to PCT/CN2017/082348 priority Critical patent/WO2018195891A1/en
Publication of WO2018195891A1 publication Critical patent/WO2018195891A1/en

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis

Definitions

  • the invention belongs to the field of image processing, and in particular relates to a method and device for evaluating non-reference image quality.
  • SCI Screen Content Image
  • image distortion may occur in the acquisition, compression, and transmission of screen content images.
  • compression processing of screen content images may cause compression distortion and reduce image visual quality;
  • bit errors random loss of image content may result in image distortion. The above distortion will affect the image quality of the screen content image, which will affect the user experience.
  • a method for solving image distortion is a quality evaluation method for a screen content image.
  • the image quality evaluation can be divided into a subjective evaluation method and an objective evaluation method.
  • the objective evaluation method includes no reference method (No Reference).
  • the current non-reference method is mainly for natural images, and natural images only contain image information.
  • screen content images contain more multimedia forms due to their diverse multimedia forms. For example, text and images in a screen content image are different from the same user's visual perception. If the existing non-reference method is used for the screen content image, the quality evaluation of the text portion may be inaccurate, and the accuracy of the evaluation result of the entire screen content image is not high.
  • the invention provides a non-reference image quality evaluation method and device, aiming at solving the problem that the accuracy of the evaluation result of the entire screen content image is low due to the use of the existing objective evaluation method for the screen content image.
  • the invention provides a non-reference image quality evaluation method, comprising: identifying a text area and an image area in a distortion image to be evaluated, and dividing the text area and the image area in the distortion image; respectively extracting the text a visually perceptual feature in the region and the image region; performing distortion classification discrimination on the visual perceptual feature from the text region by a preset classification mechanism to obtain a quality score of the text region, and a preset classification mechanism performs distortion classification and discrimination on the visual perception characteristic in the image region to obtain a quality score of the image region; and fits a quality score of the text region and a quality score of the image region And obtaining a quality score of the distorted image.
  • the invention provides a non-reference image quality evaluation apparatus, comprising: an identification module, configured to identify a text area and an image area in a distortion image to be evaluated, and divide the text area and the image area in the distortion image; An extracting module, configured to respectively extract the visually aware features in the text area and the image area; and a discriminating module, configured to perform distortion classification and discrimination on the visual sensing features in the text area by using a preset classification mechanism Obtaining a quality score of the text area, and performing distortion classification and discriminating on the visual perception characteristic in the image area by the preset classification mechanism to obtain a quality score of the image area; a fitting module, And fitting a quality score of the text area and a quality score of the image area to obtain a quality score of the distortion image.
  • the method and device for evaluating non-reference image quality identifies a text region and an image region in a distorted image to be evaluated, and divides the text region and the image region in the distorted image, respectively extracting the text region and the image region.
  • the visual perceptual feature is characterized by performing a distortion classification and discriminating on the visual perceptual feature in the text region by a preset classification mechanism, obtaining a quality score of the text region, and, by using the preset classification mechanism, the image region in the image region
  • the visual perception characteristic performs distortion classification and discrimination, obtains a quality score of the image region, and fits the quality score of the text region and the quality score of the image region to obtain a quality score of the distortion image, such that the text region and the image are obtained by
  • the quality scores are calculated separately for each region, and each can obtain an accurate quality score, and finally fits the quality score of the entire image, which can effectively improve the accuracy of the quality evaluation of the screen content image.
  • FIG. 1 is a schematic flowchart of an implementation process of a non-reference image quality evaluation method according to a first embodiment of the present invention
  • FIG. 2 is a schematic flowchart showing an implementation process of a non-reference image quality evaluation method according to a second embodiment of the present invention
  • FIG. 3 is a schematic structural diagram of a non-reference image quality evaluation apparatus according to a third embodiment of the present invention.
  • FIG. 4 is a schematic structural diagram of a non-reference image quality evaluation apparatus according to a fourth embodiment of the present invention.
  • FIG. 1 is a schematic flowchart of an implementation process of a non-reference image quality evaluation method according to a first embodiment of the present invention, which can be applied to an electronic device having an image processing function, such as a computer, and the non-reference image quality shown in FIG.
  • the evaluation method mainly includes the following steps:
  • the distorted image to be evaluated is a screen content image, that is, an image including a text area and an image area.
  • the text area of the distorted image can be known by an algorithm of text detection and recognition, and the area other than the text area is an image area, and then the text area and the image area are divided in the distorted image.
  • the text area and the visual perceptual features extracted by the image area are the same perceptual feature.
  • the method of extracting visual perception features is not limited, and visual perception features can be extracted through deep learning methods such as deep neural network, convolutional neural network, deep belief network and recurrent neural network.
  • S103 Perform distortion classification and discrimination on the visual perception feature in the text region by using a preset classification mechanism to obtain a quality score of the text region, and use the preset classification mechanism to perform the visual perception in the image region.
  • the feature performs distortion classification and discrimination to obtain the quality score of the image region.
  • the classification mechanism is preset by using a machine learning method, and the preset classification mechanism is used for performing distortion classification and discrimination on the extracted visual perception features.
  • the machine learning method used in the present invention is a support vector machine (SVM, Support). Vector Machine) and Support Vector Regression (SVR).
  • the method of fitting is not limited, and the quality score of the text region and the quality score of the image region may be fitted by a deep learning method, or the quality score of the text region and the quality score of the image region may be obtained by the SVR algorithm. Perform the fitting.
  • the text area and the image area in the distortion image to be evaluated are identified, and the text area and the image area in the distortion image are divided, and the visual perception features in the text area and the image area are respectively extracted, and a classification mechanism for performing distortion classification and discrimination on the visual perception feature in the text region, obtaining a quality score of the text region, and performing distortion classification on the visual perception characteristic in the image region by the preset classification mechanism Determining, obtaining a quality score of the image region, fitting a quality score of the text region and a quality score of the image region, obtaining a quality score of the distorted image, thereby calculating a quality score by separating the text region from the image region, Each can get an accurate quality score, and finally fit the quality score of the entire image, which can effectively improve the accuracy of the quality evaluation of the screen content image.
  • FIG. 2 is a schematic flowchart of an implementation process of a non-reference image quality evaluation method according to a second embodiment of the present invention, which can be applied to an electronic device having an image processing function, such as a computer, and the non-reference image shown in FIG.
  • the quality evaluation method mainly includes the following steps:
  • the distorted image to be evaluated is a screen content image, that is, an image including a text area and an image area.
  • the text area of the distorted image can be known by an algorithm of text detection and recognition, and the area other than the text area is an image area, and then the text area and the image area are divided in the distorted image.
  • the visual perception feature is a perceptual feature of natural scene statistics.
  • the text area and the visual perceptual features extracted by the image area are the same perceptual feature.
  • the method of extracting visual perception features is not limited, and visual perception features can be extracted through deep learning methods such as deep neural network, convolutional neural network, deep belief network and recurrent neural network.
  • the SVM algorithm and the SVR algorithm are used to train each type of the distortion type image model to obtain a quality score model of the distortion type.
  • the prediction result of the quality type of the distortion type is the quality score of the distortion type. .
  • the classification model of the distortion type obtained by the training and the quality fraction model of the distortion type are used as a preset classification mechanism.
  • the classification model of the distortion type is used to discriminate which type of distortion type the visual perception feature belongs to, and the prediction result of the classification model of the distortion type does not specifically belong to which distortion type, but is the same dimension as the distortion type number.
  • the value in each dimension is a probability value predicted to belong to the type of distortion, in other words, the prediction result of the classification model of the distortion type is a probability value of the distortion type; training one for each type of distortion image according to the distortion type
  • a quality fraction model of the distortion type the quality fraction model of the distortion type is used to predict a quality score value belonging to the distortion type, wherein the number of the distortion types is the same as the number of quality fraction models of the distortion type.
  • S206 Perform, by using the preset classification mechanism, performing distortion classification and discrimination on the visual perception feature in the text region, obtaining a quality score of the text region, and using the preset classification mechanism to perform the vision in the image region.
  • the perceptual characteristics are subjected to distortion classification and discrimination, and the quality score of the image region is obtained.
  • the visually aware feature in the text region is subjected to distortion classification and discrimination by a preset classification mechanism, and the quality score of the text region is obtained as follows:
  • a dot product between the probability value of the first distortion type and the quality score value of the first distortion type is calculated, and a quality score of the text region is calculated.
  • the first distortion type may have one or more. If there are multiple first distortion types, the probability value of each of the first distortion types and the quality belonging to the first distortion type are calculated when calculating the dot product operation. The point product is divided into numerical values to obtain the quality score of the text region.
  • the visual perception characteristic in the image region is subjected to distortion classification and discrimination by the preset classification mechanism, and the quality score of the image region is obtained as follows:
  • a dot product between the probability value of the second distortion type and the quality score value of the second distortion type is calculated, and a quality score of the image region is calculated.
  • the second distortion type may have one or more. If there are multiple second distortion types, the probability value of each of the second distortion types is equal to the quality belonging to the second distortion type when calculating the dot product operation.
  • the point product is divided into numerical values to obtain the quality score of the image region.
  • the method of fitting is not limited, and the quality score of the text region and the quality score of the image region may be fitted by a deep learning method, or the quality score of the text region and the quality score of the image region may be obtained by the SVR algorithm. Perform the fitting.
  • the text area and the image area in the distortion image to be evaluated are identified, and the text area and the image area in the distortion image are divided, and the visual perception features in the text area and the image area are respectively extracted, and a classification mechanism for performing distortion classification and discrimination on the visual perception feature in the text region, obtaining a quality score of the text region, and performing distortion classification on the visual perception characteristic in the image region by the preset classification mechanism Determining, obtaining a quality score of the image region, fitting a quality score of the text region and a quality score of the image region, obtaining a quality score of the distorted image, thereby calculating a quality score by separating the text region from the image region, Each can get an accurate quality score, and finally fit the quality score of the entire image, which can effectively improve the accuracy of the quality evaluation of the screen content image.
  • FIG. 3 is a schematic structural diagram of a non-reference image quality evaluation apparatus according to a third embodiment of the present invention. For convenience of description, only parts related to the embodiment of the present invention are shown.
  • the non-reference image quality evaluation apparatus exemplified in FIG. 3 may be an execution subject of the non-reference image quality evaluation method provided by the foregoing embodiment shown in FIGS. 1 and 2.
  • the non-reference image quality evaluation apparatus illustrated in FIG. 3 mainly includes an identification module 301, an extraction module 302, a discrimination module 303, and a fitting module 304.
  • the above functional modules are described in detail as follows:
  • the identification module 301 is configured to identify a text area and an image area in the distortion image to be evaluated, and divide the text area and the image area in the distortion image;
  • the extracting module 302 is configured to respectively extract the text region and the visually aware feature in the image region;
  • the discriminating module 303 is configured to perform distortion classification and discrimination on the visual sensing feature in the text region by using a preset classification mechanism, to obtain a quality score of the text region, and to use the preset classification mechanism in the image region.
  • the visual perception characteristic is subjected to distortion classification and discrimination to obtain a quality score of the image region;
  • the fitting module 304 is configured to fit the quality score of the text region and the quality score of the image region to obtain a quality score of the distorted image.
  • the distorted image to be evaluated is a screen content image, that is, an image including a text area and an image area.
  • the text area of the distorted image can be known by an algorithm of text detection and recognition, and the area other than the text area is an image area, and then the text area and the image area are divided in the distorted image.
  • the method of fitting is not limited, and the quality score of the text region and the quality score of the image region may be fitted by a deep learning method, or the quality score of the text region and the quality score of the image region may be obtained by the SVR algorithm. Perform the fitting.
  • each functional module is merely an example, and the actual application may be required according to requirements, such as corresponding hardware configuration requirements or software implementation convenience. It is considered that the above-mentioned function assignment is completed by different functional modules, that is, the internal structure of the above device is divided into different functional modules to complete all or part of the functions described above. Moreover, in practical applications, the corresponding functional modules in this embodiment may be implemented by corresponding hardware, or may be executed by corresponding hardware to execute corresponding software. The above description principles may be applied to various embodiments provided in this specification, and are not described herein again.
  • the identification module 301 identifies a text area and an image area in the distortion image to be evaluated, and divides the text area and the image area in the distortion image, and the extraction module 302 extracts the text area and the image area respectively.
  • the visual perception feature, the discriminating module 303 performs distortion classification and discrimination on the visual perceptual feature in the text region by using a preset classification mechanism to obtain a quality score of the text region, and the image region is determined by the preset classification mechanism.
  • the visual perceptual characteristic is subjected to distortion classification and discrimination, and a quality score of the image region is obtained, and the fitting module 304 fits the quality score of the text region and the quality score of the image region to obtain a quality score of the distorted image, such that By separating the text area and the image area to calculate the quality score, each can obtain an accurate quality score, and finally fit the quality score of the entire image, which can effectively improve the accuracy of the quality evaluation of the screen content image.
  • FIG. 4 is a schematic structural diagram of a non-reference image quality evaluation apparatus according to a fourth embodiment of the present invention.
  • the non-reference image quality evaluation apparatus exemplified in FIG. 4 may be an execution subject of the non-reference image quality evaluation method provided by the foregoing embodiment shown in FIGS. 1 and 2.
  • the reference-free image quality evaluation apparatus illustrated in FIG. 4 mainly includes an identification module 401, an extraction module 402, a training module 403, a discrimination module 404, and a fitting module 405.
  • the determination module 404 includes: a classification sub-module 4041 and a calculation sub-module 4042. .
  • the above functional modules are described in detail as follows:
  • the identification module 401 is configured to identify a text area and an image area in the distortion image to be evaluated, and divide the text area and the image area in the distortion image.
  • the distorted image to be evaluated is a screen content image, that is, an image including a text area and an image area.
  • the recognition module 401 can learn the text area of the distorted image through an algorithm of text detection and recognition, and the area other than the text area is an image area, and then the text area and the image area are divided in the distorted image.
  • the extracting module 402 is configured to respectively extract the text region and the visually perceived features in the image region.
  • the visual perception feature is a perceptual feature of natural scene statistics.
  • the text area and the visual perceptual features extracted by the image area are the same perceptual feature.
  • the manner in which the extraction module 402 extracts the visual perception features is not limited, and the extraction module 402 can extract the visual perception features through deep learning methods such as deep neural networks, convolutional neural networks, deep belief networks, and recurrent neural networks.
  • the device further includes: a training module 403;
  • a training module 403 configured to train a classification model of a distortion type by using a support vector machine SVM algorithm and a support vector regression SVR algorithm, where a prediction result of the classification model of the distortion type is a probability value of the distortion type;
  • the training module 403 is further configured to train each type of the image model of the distortion type by using the SVM algorithm and the SVR algorithm to obtain a quality score model of the distortion type, and the prediction result of the quality score model of the distortion type is The quality score of the distortion type;
  • the training module 403 is further configured to use the classification model of the distortion type obtained by the training and the quality score model of the distortion type as the preset classification mechanism.
  • the classification model of the distortion type is used to discriminate which type of distortion type the visual perception feature belongs to, and the prediction result of the classification model of the distortion type does not specifically belong to which distortion type, but is the same dimension as the distortion type number.
  • the vector, the value in each dimension is the probability value predicted to belong to the type of distortion, in other words, the prediction result of the classification model of the distortion type is the probability value of the distortion type; the training module 403 corrects each type of distortion according to the distortion type.
  • the image trains a quality score model of the distortion type, the quality score model of the distortion type being used to predict a quality score value belonging to the distortion type, wherein the number of the distortion types is the same as the number of quality score models of the distortion type.
  • a discriminating module 404 configured to perform distortion classification and discrimination on the visual perceptual feature in the text region by using a preset classification mechanism, to obtain a quality score of the text region, and to use the preset classification mechanism in the image region
  • the visual perception characteristic is subjected to distortion classification discrimination to obtain a quality score of the image region.
  • the discriminating module 404 includes: a sub-module 4041 and a computing sub-module 4042;
  • a classification sub-module 4041 configured to compare the visually-aware feature in the text region with the classification model of the distortion type, to obtain a probability value of the first distortion type to which the visual sensing feature belongs in the text region;
  • the sub-module 4041 is further configured to compare the visual perceptual feature in the text region with the quality-score model of the distortion type to obtain a quality score value of the first distortion type to which the visual perceptual feature belongs in the text region. ;
  • the calculation sub-module 4042 is configured to calculate a dot product between the probability value of the first distortion type and the quality score value of the first distortion type, and calculate a quality score of the text region.
  • the first distortion type may have one or more. If there are multiple first distortion types, the calculation sub-module 4042 calculates the probability value of each of the first distortion types and belongs to the first The quality score value of the distortion type is subjected to a dot product operation to obtain a quality score of the text region.
  • the classification sub-module 4041 is further configured to compare the visually-aware feature in the image region with the classification model of the distortion type to obtain a probability value of the second distortion type to which the visual sensing feature belongs in the image region;
  • the classification sub-module 4041 is further configured to compare the visually-aware feature in the image region with the quality-score model of the distortion type, to obtain a quality score value of the second distortion type to which the visual-aware feature in the image region belongs ;
  • the calculation sub-module 4042 is further configured to calculate a dot product between the probability value of the second distortion type and the quality score value of the second distortion type, and calculate a quality score of the image region.
  • the second distortion type may have one or more. If there are multiple second distortion types, the calculation sub-module 4042 calculates the probability of each of the second distortion types and belongs to the second when calculating the dot product operation. The quality score of the distortion type is subjected to a dot product operation to obtain a quality score of the image region.
  • the fitting module 405 is configured to fit the quality score of the text region and the quality score of the image region to obtain a quality score of the distorted image.
  • the manner of fitting is not limited, and the fitting module 405 may fit the quality score of the text region and the quality score of the image region by a deep learning method, and the fitting module 405 may also perform the quality of the text region through the SVR algorithm. The score is fitted to the quality score of the image area.
  • the identification module 401 identifies the text area and the image area in the distortion image to be evaluated, and divides the text area and the image area in the distortion image, and the extraction module 402 extracts the text area and the image area respectively.
  • the visual sensing feature, the discriminating module 403 performs distortion classification and discrimination on the visual sensing feature in the text region by using a preset classification mechanism to obtain a quality score of the text region, and the image region is determined by the preset classification mechanism.
  • the visual perceptual characteristic is subjected to distortion classification and discriminating to obtain a quality score of the image region, and the fitting module 404 fits the quality score of the text region and the quality score of the image region to obtain a quality score of the distorted image, such that By separating the text area and the image area to calculate the quality score, each can obtain an accurate quality score, and finally fit the quality score of the entire image, which can effectively improve the accuracy of the quality evaluation of the screen content image.
  • the disclosed systems, devices, and methods may be implemented in other ways.
  • the device embodiments described above are merely illustrative.
  • the division of the modules is only a logical function division.
  • there may be another division manner for example, multiple modules or components may be combined or Can be integrated into another system, or some features can be ignored or not executed.
  • the mutual coupling or direct coupling or communication link shown or discussed may be an indirect coupling or communication link through some interface, device or module, and may be electrical, mechanical or otherwise.
  • the modules described as separate components may or may not be physically separated.
  • the components displayed as modules may or may not be physical modules, that is, may be located in one place, or may be distributed to multiple network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of the embodiment.
  • each functional module in each embodiment of the present invention may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.
  • the above integrated modules can be implemented in the form of hardware or in the form of software functional modules.
  • the integrated modules if implemented in the form of software functional modules and sold or used as separate products, may be stored in a computer readable storage medium.
  • the technical solution of the present invention which is essential or contributes to the prior art, or all or part of the technical solution, may be embodied in the form of a software product stored in a storage medium.
  • a number of instructions are included to cause a computer device (which may be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods described in various embodiments of the present invention.
  • the foregoing storage medium includes: a U disk, a mobile hard disk, a read only memory (ROM, Read-Only) Memory, random access memory (RAM), disk or optical disk, and other media that can store program code.

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Image Analysis (AREA)

Abstract

A method and apparatus for evaluating quality of a non-reference image. The method comprises: identifying a text region and an image region in a distorted image to be evaluated, and dividing the text region and the image region in the distorted image (S101); respectively extracting visual perception features in the text region and in the image region (S102); performing distortion classification discrimination on the visual perception feature in the text region by means of a pre-set classification mechanism, so as to obtain a mass fraction of the text region, and performing distortion classification discrimination on the visual perception feature in the image region by means of the pre-set classification mechanism, so as to obtain a mass fraction of the image region (S103); and performing fitting on the mass fraction of the text region and the mass fraction of the image region, so as to obtain a mass fraction of the distorted image (S104). In this way, by means of separately calculating mass fractions of a text region and an image region, an accurate mass fraction can be obtained for each of the two regions, and finally, by means of fitting the mass fractions into a mass fraction of the whole image, the accuracy of quality evaluation of a screen content image can be effectively improved.

Description

无参考图像质量评价方法及装置  No reference image quality evaluation method and device 技术领域Technical field
本发明属于图像处理领域，尤其涉及一种无参考图像质量评价方法及装置。The invention belongs to the field of image processing, and in particular relates to a method and device for evaluating non-reference image quality.
背景技术Background technique
随着计算机和移动互联网技术的快速发展，多屏互动、视频游戏、远程教育等多客户端通信***得到了飞速的发展。在这类***中，各类终端之间可以相互通信，以实现屏幕内容图像 (SCI ，Screen Content Image) （终端设备绘制并显示在屏幕上的视觉内容，包括网页、邮件、地图、动画、文档和图像等）的分发和处理。在实时多客户端通信***中，屏幕内容图像在获取、压缩和传输等环节均可能出现图像失真的情况，例如，屏幕内容图像的压缩处理会造成压缩失真，降低图像的视觉质量；在传输过程中，比特位错误会导致图像内容的随机丢失，进而造成图像失真。以上失真均会影响屏幕内容图像的画质，进而影响到用户体验。With the rapid development of computer and mobile Internet technologies, multi-client communication systems such as multi-screen interaction, video games, and distance education have been rapidly developed. In this type of system, various types of terminals can communicate with each other to achieve screen content images. (SCI, Screen Content Image) (Distribution and processing of visual content that the terminal device draws and displays on the screen, including web pages, mail, maps, animations, documents, images, etc.). In real-time multi-client communication systems, image distortion may occur in the acquisition, compression, and transmission of screen content images. For example, compression processing of screen content images may cause compression distortion and reduce image visual quality; In the case of bit errors, random loss of image content may result in image distortion. The above distortion will affect the image quality of the screen content image, which will affect the user experience.
现有技术中，解决图像失真的方法为屏幕内容图像的质量评价方法。图像质量评价从方法上可以分为主观评价方法和客观评价方法。客观评价方法中包含无参考方法(No reference)。当前的无参考方法主是针对自然图像，自然图像仅包含图像信息，然而与自然图像相比，屏幕内容图像包含的多媒体形式更多，由于其多媒体形式的多样。例如，屏幕内容图像中文字和图片对于同一用户的视觉感知不同。如果对屏幕内容图像使用现有的无参考方法，会导致文字部分的质量评价不准确，进而造成整个屏幕内容图像的评价结果准确率不高。In the prior art, a method for solving image distortion is a quality evaluation method for a screen content image. The image quality evaluation can be divided into a subjective evaluation method and an objective evaluation method. The objective evaluation method includes no reference method (No Reference). The current non-reference method is mainly for natural images, and natural images only contain image information. However, compared with natural images, screen content images contain more multimedia forms due to their diverse multimedia forms. For example, text and images in a screen content image are different from the same user's visual perception. If the existing non-reference method is used for the screen content image, the quality evaluation of the text portion may be inaccurate, and the accuracy of the evaluation result of the entire screen content image is not high.
技术问题technical problem
本发明提供一种无参考图像质量评价方法及装置，旨在解决因对屏幕内容图像使用现有的客观评价方法，造成整个屏幕内容图像的评价结果准确率低的问题。 The invention provides a non-reference image quality evaluation method and device, aiming at solving the problem that the accuracy of the evaluation result of the entire screen content image is low due to the use of the existing objective evaluation method for the screen content image.
技术解决方案Technical solution
本发明提供的一种无参考图像质量评价方法，包括：识别待评价的失真图像中文本区域和图像区域，并划分所述失真图像中所述文本区域和所述图像区域；分别提取所述文本区域和所述图像区域中的视觉感知特征；通过预置的分类机制对从所述文本区域中的所述视觉感知特征进行失真分类判别，得到所述文本区域的质量分数，以及，通过所述预置的分类机制对所述图像区域中的所述视觉感知特性进行失真分类判别，得到所述图像区域的质量分数；对所述文本区域的质量分数和所述图像区域的质量分数进行拟合，得到所述失真图像的质量分数。The invention provides a non-reference image quality evaluation method, comprising: identifying a text area and an image area in a distortion image to be evaluated, and dividing the text area and the image area in the distortion image; respectively extracting the text a visually perceptual feature in the region and the image region; performing distortion classification discrimination on the visual perceptual feature from the text region by a preset classification mechanism to obtain a quality score of the text region, and a preset classification mechanism performs distortion classification and discrimination on the visual perception characteristic in the image region to obtain a quality score of the image region; and fits a quality score of the text region and a quality score of the image region And obtaining a quality score of the distorted image.
本发明提供的一种无参考图像质量评价装置，包括：识别模块，用于识别待评价的失真图像中文本区域和图像区域，并划分所述失真图像中所述文本区域和所述图像区域；提取模块，用于分别提取所述文本区域和所述图像区域中的视觉感知特征；判别模块，用于通过预置的分类机制对从所述文本区域中的所述视觉感知特征进行失真分类判别，得到所述文本区域的质量分数，以及，通过所述预置的分类机制对所述图像区域中的所述视觉感知特性进行失真分类判别，得到所述图像区域的质量分数；拟合模块，用于对所述文本区域的质量分数和所述图像区域的质量分数进行拟合，得到所述失真图像的质量分数。 The invention provides a non-reference image quality evaluation apparatus, comprising: an identification module, configured to identify a text area and an image area in a distortion image to be evaluated, and divide the text area and the image area in the distortion image; An extracting module, configured to respectively extract the visually aware features in the text area and the image area; and a discriminating module, configured to perform distortion classification and discrimination on the visual sensing features in the text area by using a preset classification mechanism Obtaining a quality score of the text area, and performing distortion classification and discriminating on the visual perception characteristic in the image area by the preset classification mechanism to obtain a quality score of the image area; a fitting module, And fitting a quality score of the text area and a quality score of the image area to obtain a quality score of the distortion image.
有益效果Beneficial effect
本发明提供的无参考图像质量评价方法及装置，识别待评价的失真图像中文本区域和图像区域，并划分该失真图像中该文本区域和该图像区域，分别提取该文本区域和该图像区域中的视觉感知特征，通过预置的分类机制对从该文本区域中的该视觉感知特征进行失真分类判别，得到该文本区域的质量分数，以及，通过该预置的分类机制对该图像区域中的该视觉感知特性进行失真分类判别，得到该图像区域的质量分数，对该文本区域的质量分数和该图像区域的质量分数进行拟合，得到该失真图像的质量分数，这样通过将文本区域和图像区域分开算出质量分数，每个均可以得到准确的质量分数，最后再拟合成整个图像的质量分数，可以有效的提高屏幕内容图像的质量评价的准确性。The method and device for evaluating non-reference image quality provided by the present invention identifies a text region and an image region in a distorted image to be evaluated, and divides the text region and the image region in the distorted image, respectively extracting the text region and the image region. The visual perceptual feature is characterized by performing a distortion classification and discriminating on the visual perceptual feature in the text region by a preset classification mechanism, obtaining a quality score of the text region, and, by using the preset classification mechanism, the image region in the image region The visual perception characteristic performs distortion classification and discrimination, obtains a quality score of the image region, and fits the quality score of the text region and the quality score of the image region to obtain a quality score of the distortion image, such that the text region and the image are obtained by The quality scores are calculated separately for each region, and each can obtain an accurate quality score, and finally fits the quality score of the entire image, which can effectively improve the accuracy of the quality evaluation of the screen content image.
附图说明DRAWINGS
为了更清楚地说明本发明实施例或现有技术中的技术方案，下面将对实施例或现有技术描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本发明的一些实施例。In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the embodiments or the description of the prior art will be briefly described below. Obviously, the drawings in the following description are only It is some embodiments of the invention.
图1是本发明第一实施例提供的无参考图像质量评价方法的实现流程示意图；1 is a schematic flowchart of an implementation process of a non-reference image quality evaluation method according to a first embodiment of the present invention;
图2是本发明第二实施例提供的无参考图像质量评价方法的实现流程示意图；2 is a schematic flowchart showing an implementation process of a non-reference image quality evaluation method according to a second embodiment of the present invention;
图3是本发明第三实施例提供的无参考图像质量评价装置的结构示意图；3 is a schematic structural diagram of a non-reference image quality evaluation apparatus according to a third embodiment of the present invention;
图4是本发明第四实施例提供的无参考图像质量评价装置的结构示意图。4 is a schematic structural diagram of a non-reference image quality evaluation apparatus according to a fourth embodiment of the present invention.
本发明的实施方式Embodiments of the invention
为使得本发明的发明目的、特征、优点能够更加的明显和易懂，下面将结合本发明实施例中的附图，对本发明实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例仅仅是本发明一部分实施例，而非全部实施例。基于本发明中的实施例，本领域技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都属于本发明保护的范围。The technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. The embodiments are merely a part of the embodiments of the invention, and not all of the embodiments. All other embodiments obtained by a person skilled in the art based on the embodiments of the present invention without creative efforts are within the scope of the present invention.
请参阅图1，图1为本发明第一实施例提供无参考图像质量评价方法的实现流程示意图，可应用于具有图像处理功能的电子设备中，如计算机，图1所示的无参考图像质量评价方法，主要包括以下步骤：Please refer to FIG. 1. FIG. 1 is a schematic flowchart of an implementation process of a non-reference image quality evaluation method according to a first embodiment of the present invention, which can be applied to an electronic device having an image processing function, such as a computer, and the non-reference image quality shown in FIG. The evaluation method mainly includes the following steps:
S101、识别待评价的失真图像中文本区域和图像区域，并划分该失真图像中该文本区域和该图像区域。S101. Identify a text area and an image area in the distortion image to be evaluated, and divide the text area and the image area in the distortion image.
该待评价的失真图像为屏幕内容图像，即包含文本区域和图像区域的图像。通过文本检测和识别的算法可以获知该失真图像的文本区域，除了该文本区域之外的区域为图像区域，然后在该失真图像中划分出该文本区域和该图像区域。The distorted image to be evaluated is a screen content image, that is, an image including a text area and an image area. The text area of the distorted image can be known by an algorithm of text detection and recognition, and the area other than the text area is an image area, and then the text area and the image area are divided in the distorted image.
S102、分别提取该文本区域和该图像区域中的视觉感知特征。S102. Extract the visual region in the text region and the image region respectively.
该文本区域和该图像区域提取的视觉感知特征为同一种感知特征。提取视觉感知特征的方式不做限定，可以通过深度学习方法，如深度神经网络、卷积神经网络、深度信念网络和递归神经网络，提取视觉感知特征。The text area and the visual perceptual features extracted by the image area are the same perceptual feature. The method of extracting visual perception features is not limited, and visual perception features can be extracted through deep learning methods such as deep neural network, convolutional neural network, deep belief network and recurrent neural network.
S103、通过预置的分类机制对从该文本区域中的该视觉感知特征进行失真分类判别，得到该文本区域的质量分数，以及，通过该预置的分类机制对该图像区域中的该视觉感知特性进行失真分类判别，得到该图像区域的质量分数。S103. Perform distortion classification and discrimination on the visual perception feature in the text region by using a preset classification mechanism to obtain a quality score of the text region, and use the preset classification mechanism to perform the visual perception in the image region. The feature performs distortion classification and discrimination to obtain the quality score of the image region.
采用机器学习方式预先设置分类机制，该预置的分类机制用于对提取的视觉感知特征的进行失真分类判别。本发明所使用的机器学习方式为支持向量机（SVM，Support Vector Machine）以及支持向量回归(SVR，Support Vector Regression)。 The classification mechanism is preset by using a machine learning method, and the preset classification mechanism is used for performing distortion classification and discrimination on the extracted visual perception features. The machine learning method used in the present invention is a support vector machine (SVM, Support). Vector Machine) and Support Vector Regression (SVR).
S104、对该文本区域的质量分数和该图像区域的质量分数进行拟合，得到该失真图像的质量分数。S104. Fit the quality score of the text region and the quality score of the image region to obtain a quality score of the distorted image.
拟合的方式不做限定，可以通过深度学习方法对该文本区域的质量分数和该图像区域的质量分数进行拟合，也可以通过SVR算法对该文本区域的质量分数和该图像区域的质量分数进行拟合。The method of fitting is not limited, and the quality score of the text region and the quality score of the image region may be fitted by a deep learning method, or the quality score of the text region and the quality score of the image region may be obtained by the SVR algorithm. Perform the fitting.
本发明实施例中，识别待评价的失真图像中文本区域和图像区域，并划分该失真图像中该文本区域和该图像区域，分别提取该文本区域和该图像区域中的视觉感知特征，通过预置的分类机制对从该文本区域中的该视觉感知特征进行失真分类判别，得到该文本区域的质量分数，以及，通过该预置的分类机制对该图像区域中的该视觉感知特性进行失真分类判别，得到该图像区域的质量分数，对该文本区域的质量分数和该图像区域的质量分数进行拟合，得到该失真图像的质量分数，这样通过将文本区域和图像区域分开算出质量分数，每个均可以得到准确的质量分数，最后再拟合成整个图像的质量分数，可以有效的提高屏幕内容图像的质量评价的准确性。In the embodiment of the present invention, the text area and the image area in the distortion image to be evaluated are identified, and the text area and the image area in the distortion image are divided, and the visual perception features in the text area and the image area are respectively extracted, and a classification mechanism for performing distortion classification and discrimination on the visual perception feature in the text region, obtaining a quality score of the text region, and performing distortion classification on the visual perception characteristic in the image region by the preset classification mechanism Determining, obtaining a quality score of the image region, fitting a quality score of the text region and a quality score of the image region, obtaining a quality score of the distorted image, thereby calculating a quality score by separating the text region from the image region, Each can get an accurate quality score, and finally fit the quality score of the entire image, which can effectively improve the accuracy of the quality evaluation of the screen content image.
请参阅图2，图2为本发明第二实施例提供的无参考图像质量评价方法的实现流程示意图，可应用于具有图像处理功能的电子设备中，如计算机，图2所示的无参考图像质量评价方法，主要包括以下步骤：Please refer to FIG. 2. FIG. 2 is a schematic flowchart of an implementation process of a non-reference image quality evaluation method according to a second embodiment of the present invention, which can be applied to an electronic device having an image processing function, such as a computer, and the non-reference image shown in FIG. The quality evaluation method mainly includes the following steps:
S201、识别待评价的失真图像中文本区域和图像区域，并划分该失真图像中该文本区域和该图像区域。S201. Identify a text area and an image area in the distortion image to be evaluated, and divide the text area and the image area in the distortion image.
该待评价的失真图像为屏幕内容图像，即包含文本区域和图像区域的图像。通过文本检测和识别的算法可以获知该失真图像的文本区域，除了该文本区域之外的区域为图像区域，然后在该失真图像中划分出该文本区域和该图像区域。The distorted image to be evaluated is a screen content image, that is, an image including a text area and an image area. The text area of the distorted image can be known by an algorithm of text detection and recognition, and the area other than the text area is an image area, and then the text area and the image area are divided in the distorted image.
S202、分别提取该文本区域和该图像区域中的视觉感知特征。S202. Extract the visual region in the text region and the image region respectively.
该视觉感知特征为自然场景统计的感知特征。该文本区域和该图像区域提取的视觉感知特征为同一种感知特征。提取视觉感知特征的方式不做限定，可以通过深度学习方法，如深度神经网络、卷积神经网络、深度信念网络和递归神经网络，提取视觉感知特征。The visual perception feature is a perceptual feature of natural scene statistics. The text area and the visual perceptual features extracted by the image area are the same perceptual feature. The method of extracting visual perception features is not limited, and visual perception features can be extracted through deep learning methods such as deep neural network, convolutional neural network, deep belief network and recurrent neural network.
S203、通过SVM算法和SVR算法，训练失真类型的分类模型，该失真类型的分类模型的预测结果为该失真类型的概率值。S203. Train a classification model of the distortion type by using an SVM algorithm and an SVR algorithm, and the prediction result of the classification model of the distortion type is a probability value of the distortion type.
S204、通过该SVM算法和该SVR算法，对每一类该失真类型的图像模型进行训练，得到该失真类型的质量分数模型，该失真类型的质量分数模型的预测结果为该失真类型的质量分数。S204. The SVM algorithm and the SVR algorithm are used to train each type of the distortion type image model to obtain a quality score model of the distortion type. The prediction result of the quality type of the distortion type is the quality score of the distortion type. .
S205、将训练得到的该失真类型的分类模型以及该失真类型的质量分数模型作为预置的分类机制。S205. The classification model of the distortion type obtained by the training and the quality fraction model of the distortion type are used as a preset classification mechanism.
这里预先训练两种模型，一种为该失真类型的分类模型，另一种为该失真类型的质量分数模型。该失真类型的分类模型用于判别该视觉感知特征属于哪一种失真类型的分类模型，该失真类型的分类模型的预测结果不是具体属于哪一失真类型，而是与失真类型数同维度的特征向量，每个维度上的值就是预测属于该失真类型的概率值，换言之，该失真类型的分类模型的预测结果为该失真类型的概率值；通过按照该失真类型对每一类失真图像训练一个该失真类型的质量分数模型，该失真类型的质量分数模型用于预测属于该失真类型的质量分数值，其中该失真类型的数量与该失真类型的质量分数模型的数量相同。Here two models are pre-trained, one is the classification model of the distortion type, and the other is the quality fraction model of the distortion type. The classification model of the distortion type is used to discriminate which type of distortion type the visual perception feature belongs to, and the prediction result of the classification model of the distortion type does not specifically belong to which distortion type, but is the same dimension as the distortion type number. a vector, the value in each dimension is a probability value predicted to belong to the type of distortion, in other words, the prediction result of the classification model of the distortion type is a probability value of the distortion type; training one for each type of distortion image according to the distortion type A quality fraction model of the distortion type, the quality fraction model of the distortion type is used to predict a quality score value belonging to the distortion type, wherein the number of the distortion types is the same as the number of quality fraction models of the distortion type.
S206、通过该预置的分类机制对从该文本区域中的该视觉感知特征进行失真分类判别，得到该文本区域的质量分数，以及，通过该预置的分类机制对该图像区域中的该视觉感知特性进行失真分类判别，得到该图像区域的质量分数。S206. Perform, by using the preset classification mechanism, performing distortion classification and discrimination on the visual perception feature in the text region, obtaining a quality score of the text region, and using the preset classification mechanism to perform the vision in the image region. The perceptual characteristics are subjected to distortion classification and discrimination, and the quality score of the image region is obtained.
进一步地，该通过预置的分类机制对该文本区域中的该视觉感知特征进行失真分类判别，得到该文本区域的质量分数具体为：Further, the visually aware feature in the text region is subjected to distortion classification and discrimination by a preset classification mechanism, and the quality score of the text region is obtained as follows:
将该文本区域中的该视觉感知特征与该失真类型的分类模型相比较，得到该文本区域中的该视觉感知特征所属的第一失真类型的概率值；Comparing the visual perception feature in the text region with the classification model of the distortion type to obtain a probability value of the first distortion type to which the visual perception feature belongs in the text region;
以及，将该文本区域中的该视觉感知特征与该失真类型的质量分数模型相比较，得到该文本区域中的该视觉感知特征所属的该第一失真类型的质量分数值；And comparing the visual perception feature in the text region with the quality score model of the distortion type to obtain a quality score value of the first distortion type to which the visual perception feature belongs in the text region;
计算该第一失真类型的概率值和该第一失真类型的质量分数值之间的点积，算出该文本区域的质量分数。A dot product between the probability value of the first distortion type and the quality score value of the first distortion type is calculated, and a quality score of the text region is calculated.
该第一失真类型可以有一个或者多个，如果该第一失真类型有多个，则在计算点积运算时，每个该第一失真类型的概率值均与属于该第一失真类型的质量分数值进行点积运算，得到该文本区域的质量分数。The first distortion type may have one or more. If there are multiple first distortion types, the probability value of each of the first distortion types and the quality belonging to the first distortion type are calculated when calculating the dot product operation. The point product is divided into numerical values to obtain the quality score of the text region.
进一步地，该通过该预置的分类机制对该图像区域中的该视觉感知特性进行失真分类判别，得到该图像区域的质量分数具体为：Further, the visual perception characteristic in the image region is subjected to distortion classification and discrimination by the preset classification mechanism, and the quality score of the image region is obtained as follows:
将该图像区域中的该视觉感知特征与该失真类型的分类模型相比较，得到该图像区域中的该视觉感知特征所属的第二失真类型的概率值；Comparing the visual perception feature in the image region with the classification model of the distortion type to obtain a probability value of the second distortion type to which the visual perception feature belongs in the image region;
以及，将该图像区域中的该视觉感知特征与该失真类型的质量分数模型相比较，得到该图像区域中的该视觉感知特征所属的该第二失真类型的质量分数值；And comparing the visual perceptual feature in the image region with the quality fraction model of the distortion type to obtain a quality score value of the second distortion type to which the visual perceptual feature belongs in the image region;
计算该第二失真类型的概率值和该第二失真类型的质量分数值之间的点积，算出该图像区域的质量分数。A dot product between the probability value of the second distortion type and the quality score value of the second distortion type is calculated, and a quality score of the image region is calculated.
该第二失真类型可以有一个或者多个，如果该第二失真类型有多个，则在计算点积运算时，每个该第二失真类型的概率值均与属于该第二失真类型的质量分数值进行点积运算，得到该图像区域的质量分数。The second distortion type may have one or more. If there are multiple second distortion types, the probability value of each of the second distortion types is equal to the quality belonging to the second distortion type when calculating the dot product operation. The point product is divided into numerical values to obtain the quality score of the image region.
S207、对该文本区域的质量分数和该图像区域的质量分数进行拟合，得到该失真图像的质量分数。S207. Fit the quality score of the text area and the quality score of the image area to obtain a quality score of the distortion image.
拟合的方式不做限定，可以通过深度学习方法对该文本区域的质量分数和该图像区域的质量分数进行拟合，也可以通过SVR算法对该文本区域的质量分数和该图像区域的质量分数进行拟合。The method of fitting is not limited, and the quality score of the text region and the quality score of the image region may be fitted by a deep learning method, or the quality score of the text region and the quality score of the image region may be obtained by the SVR algorithm. Perform the fitting.
本发明实施例中，识别待评价的失真图像中文本区域和图像区域，并划分该失真图像中该文本区域和该图像区域，分别提取该文本区域和该图像区域中的视觉感知特征，通过预置的分类机制对从该文本区域中的该视觉感知特征进行失真分类判别，得到该文本区域的质量分数，以及，通过该预置的分类机制对该图像区域中的该视觉感知特性进行失真分类判别，得到该图像区域的质量分数，对该文本区域的质量分数和该图像区域的质量分数进行拟合，得到该失真图像的质量分数，这样通过将文本区域和图像区域分开算出质量分数，每个均可以得到准确的质量分数，最后再拟合成整个图像的质量分数，可以有效的提高屏幕内容图像的质量评价的准确性。In the embodiment of the present invention, the text area and the image area in the distortion image to be evaluated are identified, and the text area and the image area in the distortion image are divided, and the visual perception features in the text area and the image area are respectively extracted, and a classification mechanism for performing distortion classification and discrimination on the visual perception feature in the text region, obtaining a quality score of the text region, and performing distortion classification on the visual perception characteristic in the image region by the preset classification mechanism Determining, obtaining a quality score of the image region, fitting a quality score of the text region and a quality score of the image region, obtaining a quality score of the distorted image, thereby calculating a quality score by separating the text region from the image region, Each can get an accurate quality score, and finally fit the quality score of the entire image, which can effectively improve the accuracy of the quality evaluation of the screen content image.
请参阅图3，图3是本发明第三实施例提供的无参考图像质量评价装置的结构示意图，为了便于说明，仅示出了与本发明实施例相关的部分。图3示例的无参考图像质量评价装置可以是前述图1和图2所示实施例提供的无参考图像质量评价方法的执行主体。图3示例的无参考图像质量评价装置，主要包括：识别模块301、提取模块302、判别模块303和拟合模块304。以上各功能模块详细说明如下：Referring to FIG. 3, FIG. 3 is a schematic structural diagram of a non-reference image quality evaluation apparatus according to a third embodiment of the present invention. For convenience of description, only parts related to the embodiment of the present invention are shown. The non-reference image quality evaluation apparatus exemplified in FIG. 3 may be an execution subject of the non-reference image quality evaluation method provided by the foregoing embodiment shown in FIGS. 1 and 2. The non-reference image quality evaluation apparatus illustrated in FIG. 3 mainly includes an identification module 301, an extraction module 302, a discrimination module 303, and a fitting module 304. The above functional modules are described in detail as follows:
识别模块301，用于识别待评价的失真图像中文本区域和图像区域，并划分该失真图像中该文本区域和该图像区域；The identification module 301 is configured to identify a text area and an image area in the distortion image to be evaluated, and divide the text area and the image area in the distortion image;
提取模块302，用于分别提取该文本区域和该图像区域中的视觉感知特征；The extracting module 302 is configured to respectively extract the text region and the visually aware feature in the image region;
判别模块303，用于通过预置的分类机制对从该文本区域中的该视觉感知特征进行失真分类判别，得到该文本区域的质量分数，以及，通过该预置的分类机制对该图像区域中的该视觉感知特性进行失真分类判别，得到该图像区域的质量分数；The discriminating module 303 is configured to perform distortion classification and discrimination on the visual sensing feature in the text region by using a preset classification mechanism, to obtain a quality score of the text region, and to use the preset classification mechanism in the image region. The visual perception characteristic is subjected to distortion classification and discrimination to obtain a quality score of the image region;
拟合模块304，用于对该文本区域的质量分数和该图像区域的质量分数进行拟合，得到该失真图像的质量分数。The fitting module 304 is configured to fit the quality score of the text region and the quality score of the image region to obtain a quality score of the distorted image.
该待评价的失真图像为屏幕内容图像，即包含文本区域和图像区域的图像。通过文本检测和识别的算法可以获知该失真图像的文本区域，除了该文本区域之外的区域为图像区域，然后在该失真图像中划分出该文本区域和该图像区域。The distorted image to be evaluated is a screen content image, that is, an image including a text area and an image area. The text area of the distorted image can be known by an algorithm of text detection and recognition, and the area other than the text area is an image area, and then the text area and the image area are divided in the distorted image.
拟合的方式不做限定，可以通过深度学习方法对该文本区域的质量分数和该图像区域的质量分数进行拟合，也可以通过SVR算法对该文本区域的质量分数和该图像区域的质量分数进行拟合。The method of fitting is not limited, and the quality score of the text region and the quality score of the image region may be fitted by a deep learning method, or the quality score of the text region and the quality score of the image region may be obtained by the SVR algorithm. Perform the fitting.
本实施例未尽之细节，请参阅前述图1所示实施例的描述，此处不再赘述。For details of the embodiment, please refer to the description of the embodiment shown in FIG. 1 , and details are not described herein again.
需要说明的是，以上图3示例的无参考图像质量评价装置的实施方式中，各功能模块的划分仅是举例说明，实际应用中可以根据需要，例如相应硬件的配置要求或者软件的实现的便利考虑，而将上述功能分配由不同的功能模块完成，即将上述装置的内部结构划分成不同的功能模块，以完成以上描述的全部或者部分功能。而且，实际应用中，本实施例中的相应的功能模块可以是由相应的硬件实现，也可以由相应的硬件执行相应的软件完成。本说明书提供的各个实施例都可应用上述描述原则，以下不再赘述。It should be noted that, in the implementation manner of the non-reference image quality evaluation apparatus illustrated in FIG. 3 above, the division of each functional module is merely an example, and the actual application may be required according to requirements, such as corresponding hardware configuration requirements or software implementation convenience. It is considered that the above-mentioned function assignment is completed by different functional modules, that is, the internal structure of the above device is divided into different functional modules to complete all or part of the functions described above. Moreover, in practical applications, the corresponding functional modules in this embodiment may be implemented by corresponding hardware, or may be executed by corresponding hardware to execute corresponding software. The above description principles may be applied to various embodiments provided in this specification, and are not described herein again.
本发明实施例中，识别模块301识别待评价的失真图像中文本区域和图像区域，并划分该失真图像中该文本区域和该图像区域，提取模块302分别提取该文本区域和该图像区域中的视觉感知特征，判别模块303通过预置的分类机制对从该文本区域中的该视觉感知特征进行失真分类判别，得到该文本区域的质量分数，以及，通过该预置的分类机制对该图像区域中的该视觉感知特性进行失真分类判别，得到该图像区域的质量分数，拟合模块304对该文本区域的质量分数和该图像区域的质量分数进行拟合，得到该失真图像的质量分数，这样通过将文本区域和图像区域分开算出质量分数，每个均可以得到准确的质量分数，最后再拟合成整个图像的质量分数，可以有效的提高屏幕内容图像的质量评价的准确性。In the embodiment of the present invention, the identification module 301 identifies a text area and an image area in the distortion image to be evaluated, and divides the text area and the image area in the distortion image, and the extraction module 302 extracts the text area and the image area respectively. The visual perception feature, the discriminating module 303 performs distortion classification and discrimination on the visual perceptual feature in the text region by using a preset classification mechanism to obtain a quality score of the text region, and the image region is determined by the preset classification mechanism. The visual perceptual characteristic is subjected to distortion classification and discrimination, and a quality score of the image region is obtained, and the fitting module 304 fits the quality score of the text region and the quality score of the image region to obtain a quality score of the distorted image, such that By separating the text area and the image area to calculate the quality score, each can obtain an accurate quality score, and finally fit the quality score of the entire image, which can effectively improve the accuracy of the quality evaluation of the screen content image.
请参阅图4，图4为本发明第四实施例提供的无参考图像质量评价装置的结构示意图，为了便于说明，仅示出了与本发明实施例相关的部分。图4示例的无参考图像质量评价装置可以是前述图1和图2所示实施例提供的无参考图像质量评价方法的执行主体。图4示例的无参考图像质量评价装置，主要包括：识别模块401、提取模块402、训练模块403、判别模块404和拟合模块405，其中判别模块404包括：分类子模块4041和计算子模块4042。以上各功能模块详细说明如下：Referring to FIG. 4, FIG. 4 is a schematic structural diagram of a non-reference image quality evaluation apparatus according to a fourth embodiment of the present invention. For convenience of description, only parts related to the embodiment of the present invention are shown. The non-reference image quality evaluation apparatus exemplified in FIG. 4 may be an execution subject of the non-reference image quality evaluation method provided by the foregoing embodiment shown in FIGS. 1 and 2. The reference-free image quality evaluation apparatus illustrated in FIG. 4 mainly includes an identification module 401, an extraction module 402, a training module 403, a discrimination module 404, and a fitting module 405. The determination module 404 includes: a classification sub-module 4041 and a calculation sub-module 4042. . The above functional modules are described in detail as follows:
识别模块401，用于识别待评价的失真图像中文本区域和图像区域，并划分该失真图像中该文本区域和该图像区域。The identification module 401 is configured to identify a text area and an image area in the distortion image to be evaluated, and divide the text area and the image area in the distortion image.
该待评价的失真图像为屏幕内容图像，即包含文本区域和图像区域的图像。识别模块401通过文本检测和识别的算法可以获知该失真图像的文本区域，除了该文本区域之外的区域为图像区域，然后在该失真图像中划分出该文本区域和该图像区域。The distorted image to be evaluated is a screen content image, that is, an image including a text area and an image area. The recognition module 401 can learn the text area of the distorted image through an algorithm of text detection and recognition, and the area other than the text area is an image area, and then the text area and the image area are divided in the distorted image.
提取模块402，用于分别提取该文本区域和该图像区域中的视觉感知特征。The extracting module 402 is configured to respectively extract the text region and the visually perceived features in the image region.
该视觉感知特征为自然场景统计的感知特征。该文本区域和该图像区域提取的视觉感知特征为同一种感知特征。提取模块402提取视觉感知特征的方式不做限定，提取模块402可以通过深度学习方法，如深度神经网络、卷积神经网络、深度信念网络和递归神经网络，提取视觉感知特征。The visual perception feature is a perceptual feature of natural scene statistics. The text area and the visual perceptual features extracted by the image area are the same perceptual feature. The manner in which the extraction module 402 extracts the visual perception features is not limited, and the extraction module 402 can extract the visual perception features through deep learning methods such as deep neural networks, convolutional neural networks, deep belief networks, and recurrent neural networks.
进一步地，该装置还包括：训练模块403；Further, the device further includes: a training module 403;
训练模块403，用于通过支持向量机SVM算法和支持向量回归SVR算法，训练失真类型的分类模型，该失真类型的分类模型的预测结果为该失真类型的概率值；a training module 403, configured to train a classification model of a distortion type by using a support vector machine SVM algorithm and a support vector regression SVR algorithm, where a prediction result of the classification model of the distortion type is a probability value of the distortion type;
训练模块403，还用于通过该SVM算法和该SVR算法，对每一类该失真类型的图像模型进行训练，得到该失真类型的质量分数模型，该失真类型的质量分数模型的预测结果为该失真类型的质量分数；The training module 403 is further configured to train each type of the image model of the distortion type by using the SVM algorithm and the SVR algorithm to obtain a quality score model of the distortion type, and the prediction result of the quality score model of the distortion type is The quality score of the distortion type;
训练模块403，还用于将训练得到的所述失真类型的分类模型以及所述失真类型的质量分数模型作为所述预置的分类机制。The training module 403 is further configured to use the classification model of the distortion type obtained by the training and the quality score model of the distortion type as the preset classification mechanism.
这里预先训练两种模型，一种为该失真类型的分类模型，另一种为该失真类型的质量分数模型。该失真类型的分类模型用于判别该视觉感知特征属于哪一种失真类型的分类模型，该失真类型的分类模型的预测结果不是具体属于哪一失真类型，而是与失真类型数同维度的特征向量，每个维度上的值就是预测属于该失真类型的概率值，换言之，该失真类型的分类模型的预测结果为该失真类型的概率值；训练模块403通过按照该失真类型对每一类失真图像训练一个该失真类型的质量分数模型，该失真类型的质量分数模型用于预测属于该失真类型的质量分数值，其中该失真类型的数量与该失真类型的质量分数模型的数量相同。Here two models are pre-trained, one is the classification model of the distortion type, and the other is the quality fraction model of the distortion type. The classification model of the distortion type is used to discriminate which type of distortion type the visual perception feature belongs to, and the prediction result of the classification model of the distortion type does not specifically belong to which distortion type, but is the same dimension as the distortion type number. The vector, the value in each dimension is the probability value predicted to belong to the type of distortion, in other words, the prediction result of the classification model of the distortion type is the probability value of the distortion type; the training module 403 corrects each type of distortion according to the distortion type. The image trains a quality score model of the distortion type, the quality score model of the distortion type being used to predict a quality score value belonging to the distortion type, wherein the number of the distortion types is the same as the number of quality score models of the distortion type.
判别模块404，用于通过预置的分类机制对从该文本区域中的该视觉感知特征进行失真分类判别，得到该文本区域的质量分数，以及，通过该预置的分类机制对该图像区域中的该视觉感知特性进行失真分类判别，得到该图像区域的质量分数。a discriminating module 404, configured to perform distortion classification and discrimination on the visual perceptual feature in the text region by using a preset classification mechanism, to obtain a quality score of the text region, and to use the preset classification mechanism in the image region The visual perception characteristic is subjected to distortion classification discrimination to obtain a quality score of the image region.
进一步地，判别模块404包括：分类子模块4041和计算子模块4042；Further, the discriminating module 404 includes: a sub-module 4041 and a computing sub-module 4042;
分类子模块4041，用于将该文本区域中的该视觉感知特征与该失真类型的分类模型相比较，得到该文本区域中的该视觉感知特征所属的第一失真类型的概率值；a classification sub-module 4041, configured to compare the visually-aware feature in the text region with the classification model of the distortion type, to obtain a probability value of the first distortion type to which the visual sensing feature belongs in the text region;
分类子模块4041，还用于将该文本区域中的该视觉感知特征与该失真类型的质量分数模型相比较，得到该文本区域中的该视觉感知特征所属的该第一失真类型的质量分数值；The sub-module 4041 is further configured to compare the visual perceptual feature in the text region with the quality-score model of the distortion type to obtain a quality score value of the first distortion type to which the visual perceptual feature belongs in the text region. ;
计算子模块4042，用于计算该第一失真类型的概率值和该第一失真类型的质量分数值之间的点积，算出该文本区域的质量分数。The calculation sub-module 4042 is configured to calculate a dot product between the probability value of the first distortion type and the quality score value of the first distortion type, and calculate a quality score of the text region.
该第一失真类型可以有一个或者多个，如果该第一失真类型有多个，则计算子模块4042在计算点积运算时，每个该第一失真类型的概率值均与属于该第一失真类型的质量分数值进行点积运算，得到该文本区域的质量分数。The first distortion type may have one or more. If there are multiple first distortion types, the calculation sub-module 4042 calculates the probability value of each of the first distortion types and belongs to the first The quality score value of the distortion type is subjected to a dot product operation to obtain a quality score of the text region.
分类子模块4041，还用于将该图像区域中的该视觉感知特征与该失真类型的分类模型相比较，得到该图像区域中的该视觉感知特征所属的第二失真类型的概率值；The classification sub-module 4041 is further configured to compare the visually-aware feature in the image region with the classification model of the distortion type to obtain a probability value of the second distortion type to which the visual sensing feature belongs in the image region;
分类子模块4041，还用于将该图像区域中的该视觉感知特征与该失真类型的质量分数模型相比较，得到该图像区域中的该视觉感知特征所属的该第二失真类型的质量分数值；The classification sub-module 4041 is further configured to compare the visually-aware feature in the image region with the quality-score model of the distortion type, to obtain a quality score value of the second distortion type to which the visual-aware feature in the image region belongs ;
计算子模块4042，还用于计算该第二失真类型的概率值和该第二失真类型的质量分数值之间的点积，算出该图像区域的质量分数。The calculation sub-module 4042 is further configured to calculate a dot product between the probability value of the second distortion type and the quality score value of the second distortion type, and calculate a quality score of the image region.
该第二失真类型可以有一个或者多个，如果该第二失真类型有多个，则计算子模块4042在计算点积运算时，每个该第二失真类型的概率值均与属于该第二失真类型的质量分数值进行点积运算，得到该图像区域的质量分数。The second distortion type may have one or more. If there are multiple second distortion types, the calculation sub-module 4042 calculates the probability of each of the second distortion types and belongs to the second when calculating the dot product operation. The quality score of the distortion type is subjected to a dot product operation to obtain a quality score of the image region.
拟合模块405，用于对该文本区域的质量分数和该图像区域的质量分数进行拟合，得到该失真图像的质量分数。The fitting module 405 is configured to fit the quality score of the text region and the quality score of the image region to obtain a quality score of the distorted image.
拟合的方式不做限定，拟合模块405可以通过深度学习方法对该文本区域的质量分数和该图像区域的质量分数进行拟合，拟合模块405也可以通过SVR算法对该文本区域的质量分数和该图像区域的质量分数进行拟合。The manner of fitting is not limited, and the fitting module 405 may fit the quality score of the text region and the quality score of the image region by a deep learning method, and the fitting module 405 may also perform the quality of the text region through the SVR algorithm. The score is fitted to the quality score of the image area.
本实施例未尽之细节，请参阅前述图1和图2所示实施例的描述，此处不再赘述。For details of the embodiment, please refer to the description of the embodiment shown in FIG. 1 and FIG. 2, and details are not described herein again.
本发明实施例中，识别模块401识别待评价的失真图像中文本区域和图像区域，并划分该失真图像中该文本区域和该图像区域，提取模块402分别提取该文本区域和该图像区域中的视觉感知特征，判别模块403通过预置的分类机制对从该文本区域中的该视觉感知特征进行失真分类判别，得到该文本区域的质量分数，以及，通过该预置的分类机制对该图像区域中的该视觉感知特性进行失真分类判别，得到该图像区域的质量分数，拟合模块404对该文本区域的质量分数和该图像区域的质量分数进行拟合，得到该失真图像的质量分数，这样通过将文本区域和图像区域分开算出质量分数，每个均可以得到准确的质量分数，最后再拟合成整个图像的质量分数，可以有效的提高屏幕内容图像的质量评价的准确性。In the embodiment of the present invention, the identification module 401 identifies the text area and the image area in the distortion image to be evaluated, and divides the text area and the image area in the distortion image, and the extraction module 402 extracts the text area and the image area respectively. The visual sensing feature, the discriminating module 403 performs distortion classification and discrimination on the visual sensing feature in the text region by using a preset classification mechanism to obtain a quality score of the text region, and the image region is determined by the preset classification mechanism. The visual perceptual characteristic is subjected to distortion classification and discriminating to obtain a quality score of the image region, and the fitting module 404 fits the quality score of the text region and the quality score of the image region to obtain a quality score of the distorted image, such that By separating the text area and the image area to calculate the quality score, each can obtain an accurate quality score, and finally fit the quality score of the entire image, which can effectively improve the accuracy of the quality evaluation of the screen content image.
在本申请所提供的多个实施例中，应该理解到，所揭露的***、装置和方法，可以通过其它的方式实现。例如，以上所描述的装置实施例仅仅是示意性的，例如，所述模块的划分，仅仅为一种逻辑功能划分，实际实现时可以有另外的划分方式，例如多个模块或组件可以结合或者可以集成到另一个***，或一些特征可以忽略，或不执行。另一点，所显示或讨论的相互之间的耦合或直接耦合或通信链接可以是通过一些接口，装置或模块的间接耦合或通信链接，可以是电性，机械或其它的形式。In the various embodiments provided herein, it should be understood that the disclosed systems, devices, and methods may be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the modules is only a logical function division. In actual implementation, there may be another division manner, for example, multiple modules or components may be combined or Can be integrated into another system, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication link shown or discussed may be an indirect coupling or communication link through some interface, device or module, and may be electrical, mechanical or otherwise.
所述作为分离部件说明的模块可以是或者也可以不是物理上分开的，作为模块显示的部件可以是或者也可以不是物理模块，即可以位于一个地方，或者也可以分布到多个网络模块上。可以根据实际的需要选择其中的部分或者全部模块来实现本实施例方案的目的。The modules described as separate components may or may not be physically separated. The components displayed as modules may or may not be physical modules, that is, may be located in one place, or may be distributed to multiple network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of the embodiment.
另外，在本发明各个实施例中的各功能模块可以集成在一个处理模块中，也可以是各个模块单独物理存在，也可以两个或两个以上模块集成在一个模块中。上述集成的模块既可以采用硬件的形式实现，也可以采用软件功能模块的形式实现。In addition, each functional module in each embodiment of the present invention may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module. The above integrated modules can be implemented in the form of hardware or in the form of software functional modules.
所述集成的模块如果以软件功能模块的形式实现并作为独立的产品销售或使用时，可以存储在一个计算机可读取存储介质中。基于这样的理解，本发明的技术方案本质上或者说对现有技术做出贡献的部分或者该技术方案的全部或部分可以以软件产品的形式体现出来，该计算机软件产品存储在一个存储介质中，包括若干指令用以使得一台计算机设备（可以是个人计算机，服务器，或者网络设备等）执行本发明各个实施例所述方法的全部或部分步骤。而前述的存储介质包括：U盘、移动硬盘、只读存储器（ROM，Read-Only Memory）、随机存取存储器（RAM，Random Access Memory）、磁碟或者光盘等各种可以存储程序代码的介质。The integrated modules, if implemented in the form of software functional modules and sold or used as separate products, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present invention, which is essential or contributes to the prior art, or all or part of the technical solution, may be embodied in the form of a software product stored in a storage medium. A number of instructions are included to cause a computer device (which may be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods described in various embodiments of the present invention. The foregoing storage medium includes: a U disk, a mobile hard disk, a read only memory (ROM, Read-Only) Memory, random access memory (RAM), disk or optical disk, and other media that can store program code.
需要说明的是，对于前述的各方法实施例，为了简便描述，故将其都表述为一系列的动作组合，但是本领域技术人员应该知悉，本发明并不受所描述的动作顺序的限制，因为依据本发明，某些步骤可以采用其它顺序或者同时进行。其次，本领域技术人员也应该知悉，说明书中所描述的实施例均属于优选实施例，所涉及的动作和模块并不一定都是本发明所必须的。It should be noted that, for the foregoing method embodiments, for the sake of brevity, they are all described as a series of action combinations, but those skilled in the art should understand that the present invention is not limited by the described action sequence. Because certain steps may be performed in other sequences or concurrently in accordance with the present invention. In the following, those skilled in the art should also understand that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present invention.
在上述实施例中，对各个实施例的描述都各有侧重，某个实施例中没有详述的部分，可以参见其它实施例的相关描述。In the above embodiments, the descriptions of the various embodiments are all focused, and the parts that are not detailed in a certain embodiment can be referred to the related descriptions of other embodiments.
以上为对本发明所提供的无参考图像质量评价方法及装置的描述，对于本领域的技术人员，依据本发明实施例的思想，在具体实施方式及应用范围上均会有改变之处，综上，本说明书内容不应理解为对本发明的限制。The above is a description of the non-reference image quality evaluation method and apparatus provided by the present invention. For those skilled in the art, according to the idea of the embodiment of the present invention, there will be changes in the specific implementation manner and the application range. The contents of this specification are not to be construed as limiting the invention.

Claims (10)

  1. 一种无参考图像质量评价方法，其特征在于，包括：A non-reference image quality evaluation method, comprising:
    识别待评价的失真图像中文本区域和图像区域，并划分所述失真图像中所述文本区域和所述图像区域；Identifying a text area and an image area in the distorted image to be evaluated, and dividing the text area and the image area in the distorted image;
    分别提取所述文本区域和所述图像区域中的视觉感知特征；Extracting the visually aware features in the text area and the image area, respectively;
    通过预置的分类机制对从所述文本区域中的所述视觉感知特征进行失真分类判别，得到所述文本区域的质量分数，以及，通过所述预置的分类机制对所述图像区域中的所述视觉感知特性进行失真分类判别，得到所述图像区域的质量分数；Performing distortion classification and discrimination on the visually-aware features in the text region by a preset classification mechanism to obtain a quality score of the text region, and, by using the preset classification mechanism, in the image region Performing distortion classification discrimination on the visual perception characteristic to obtain a quality score of the image region;
    对所述文本区域的质量分数和所述图像区域的质量分数进行拟合，得到所述失真图像的质量分数。A quality score of the text region and a quality score of the image region are fitted to obtain a quality score of the distorted image.
  2. 根据权利要求1所述的方法，其特征在于，所述通过预置的分类机制对从所述文本区域中的所述视觉感知特征进行失真分类判别，得到所述文本区域的质量分数，以及，通过所述预置的分类机制对所述图像区域中的所述视觉感知特性进行失真分类判别，得到所述图像区域的质量分数之前还包括：The method according to claim 1, wherein the discriminating and discriminating the visually perceptible features in the text region by a preset classification mechanism obtains a quality score of the text region, and Performing distortion classification and discriminating on the visual perceptual characteristics in the image region by using the preset classification mechanism, and before obtaining the quality score of the image region, the method further includes:
    通过支持向量机SVM算法和支持向量回归SVR算法，训练失真类型的分类模型，所述失真类型的分类模型的预测结果为所述失真类型的概率值；Training a distortion type classification model by using a support vector machine SVM algorithm and a support vector regression SVR algorithm, wherein the prediction result of the classification model of the distortion type is a probability value of the distortion type;
    以及，通过所述SVM算法和所述SVR算法，对每一类所述失真类型的图像模型进行训练，得到所述失真类型的质量分数模型，所述失真类型的质量分数模型的预测结果为所述失真类型的质量分数；And, by using the SVM algorithm and the SVR algorithm, training each type of the distortion type image model to obtain a quality score model of the distortion type, and the prediction result of the distortion type quality score model is The quality score of the type of distortion;
    将训练得到的所述失真类型的分类模型以及所述失真类型的质量分数模型作为所述预置的分类机制。The classification model of the distortion type obtained by the training and the quality score model of the distortion type are used as the preset classification mechanism.
  3. 根据权利要求2所述的方法，其特征在于，所述通过预置的分类机制对所述文本区域中的所述视觉感知特征进行失真分类判别，得到所述文本区域的质量分数包括：The method according to claim 2, wherein the visually discriminating features in the text region are subjected to distortion classification and discrimination by a preset classification mechanism, and the quality score of the text region is obtained by:
    将所述文本区域中的所述视觉感知特征与所述失真类型的分类模型相比较，得到所述文本区域中的所述视觉感知特征所属的第一失真类型的概率值；Comparing the visually-aware feature in the text region with a classification model of the distortion type to obtain a probability value of a first distortion type to which the visually-aware feature in the text region belongs;
    以及，将所述文本区域中的所述视觉感知特征与所述失真类型的质量分数模型相比较，得到所述文本区域中的所述视觉感知特征所属的所述第一失真类型的质量分数值；And comparing the visual perceptual feature in the text region with a quality score model of the distortion type to obtain a quality score value of the first distortion type to which the visual perceptual feature in the text region belongs ;
    计算所述第一失真类型的概率值和所述第一失真类型的质量分数值之间的点积，算出所述文本区域的质量分数。A dot product between the probability value of the first distortion type and the quality score value of the first distortion type is calculated, and a quality score of the text region is calculated.
  4. 根据权利要求3所述的方法，其特征在于，所述通过所述预置的分类机制对所述图像区域中的所述视觉感知特性进行失真分类判别，得到所述图像区域的质量分数包括：The method according to claim 3, wherein the discriminating and discriminating the visual perceptual characteristics in the image region by the preset classification mechanism, and obtaining the quality score of the image region comprises:
    将所述图像区域中的所述视觉感知特征与所述失真类型的分类模型相比较，得到所述图像区域中的所述视觉感知特征所属的第二失真类型的概率值；Comparing the visually-aware feature in the image region with a classification model of the distortion type to obtain a probability value of a second distortion type to which the visually-aware feature in the image region belongs;
    以及，将所述图像区域中的所述视觉感知特征与所述失真类型的质量分数模型相比较，得到所述图像区域中的所述视觉感知特征所属的所述第二失真类型的质量分数值；And comparing the visual perceptual feature in the image region with a quality score model of the distortion type to obtain a quality score value of the second distortion type to which the visual perceptual feature in the image region belongs ;
    计算所述第二失真类型的概率值和所述第二失真类型的质量分数值之间的点积，算出所述图像区域的质量分数。A dot product between the probability value of the second distortion type and the quality score value of the second distortion type is calculated, and a quality score of the image region is calculated.
  5. 根据权利要求1-4任一项所述的方法，其特征在于，A method according to any one of claims 1 to 4, characterized in that
    所述视觉感知特征为:自然场景统计的感知特征。The visual perception feature is: a perceptual feature of natural scene statistics.
  6. 一种无参考图像质量评价装置，其特征在于，所述装置包括：A non-reference image quality evaluation apparatus, characterized in that the apparatus comprises:
    识别模块，用于识别待评价的失真图像中文本区域和图像区域，并划分所述失真图像中所述文本区域和所述图像区域；An identification module, configured to identify a text area and an image area in the distorted image to be evaluated, and divide the text area and the image area in the distorted image;
    提取模块，用于分别提取所述文本区域和所述图像区域中的视觉感知特征；An extraction module, configured to respectively extract the visually aware features in the text area and the image area;
    判别模块，用于通过预置的分类机制对从所述文本区域中的所述视觉感知特征进行失真分类判别，得到所述文本区域的质量分数，以及，通过所述预置的分类机制对所述图像区域中的所述视觉感知特性进行失真分类判别，得到所述图像区域的质量分数；a discriminating module, configured to perform distortion classification and discriminating on the visual perceptual feature in the text region by a preset classification mechanism, to obtain a quality score of the text region, and, by using the preset classification mechanism Performing distortion classification discrimination on the visual perception characteristic in the image region to obtain a quality score of the image region;
    拟合模块，用于对所述文本区域的质量分数和所述图像区域的质量分数进行拟合，得到所述失真图像的质量分数。And a fitting module, configured to fit a quality score of the text region and a quality score of the image region to obtain a quality score of the distorted image.
  7. 根据权利要求6所述的装置，其特征在于，所述装置还包括：The device according to claim 6, wherein the device further comprises:
    训练模块，用于通过SVM算法和SVR算法，训练失真类型的分类模型，所述失真类型的分类模型的预测结果为所述失真类型的概率值；a training module, configured to train a classification model of a distortion type by using an SVM algorithm and an SVR algorithm, where a prediction result of the classification model of the distortion type is a probability value of the distortion type;
    所述训练模块，还用于通过所述SVM和所述SVR，对每一类所述失真类型的图像模型进行训练，得到所述失真类型的质量分数模型，所述失真类型的质量分数模型的预测结果为所述失真类型的质量分数；The training module is further configured to train, by using the SVM and the SVR, an image model of each type of the distortion type to obtain a quality score model of the distortion type, and a quality score model of the distortion type The predicted result is the quality score of the distortion type;
    所述训练模块，还用于将训练得到的所述失真类型的分类模型以及所述失真类型的质量分数模型作为所述预置的分类机制。The training module is further configured to use the classification model of the distortion type obtained by the training and the quality score model of the distortion type as the preset classification mechanism.
  8. 根据权利要求7所述的装置，其特征在于，所述判别模块包括：The device according to claim 7, wherein the discriminating module comprises:
    分类子模块，用于将所述文本区域中的所述视觉感知特征与所述失真类型的分类模型相比较，得到所述文本区域中的所述视觉感知特征所属的第一失真类型的概率值；a classifying sub-module, configured to compare the visual perceptual feature in the text region with a classification model of the distortion type, to obtain a probability value of a first distortion type to which the visual perceptual feature belongs in the text region ;
    所述分类子模块，还用于将所述文本区域中的所述视觉感知特征与所述失真类型的质量分数模型相比较，得到所述文本区域中的所述视觉感知特征所属的所述第一失真类型的质量分数值；The classifying sub-module is further configured to compare the visually-aware feature in the text region with a quality-score model of the distortion type, to obtain the number of the visual-aware feature in the text region a quality score value of a distortion type;
    计算子模块，用于计算所述第一失真类型的概率值和所述第一失真类型的质量分数值之间的点积，算出所述文本区域的质量分数。And a calculation submodule configured to calculate a dot product between the probability value of the first distortion type and the quality score value of the first distortion type, and calculate a quality score of the text region.
  9. 根据权利要求8所述的装置，其特征在于， The device of claim 8 wherein:
    所述分类子模块，还用于将所述图像区域中的所述视觉感知特征与所述失真类型的分类模型相比较，得到所述图像区域中的所述视觉感知特征所属的第二失真类型的概率值；The classifying sub-module is further configured to compare the visually-aware feature in the image region with a classification model of the distortion type to obtain a second distortion type to which the visual-aware feature in the image region belongs Probability value;
    所述分类子模块，还用于将所述图像区域中的所述视觉感知特征与所述失真类型的质量分数模型相比较，得到所述图像区域中的所述视觉感知特征所属的所述第二失真类型的质量分数值；The classifying sub-module is further configured to compare the visually-aware feature in the image region with a quality-score model of the distortion type to obtain the first part of the image-aware feature in the image region The mass fraction value of the second distortion type;
    所述计算子模块，还用于计算所述第二失真类型的概率值和所述第二失真类型的质量分数值之间的点积，算出所述图像区域的质量分数。The calculation submodule is further configured to calculate a dot product between the probability value of the second distortion type and the quality score value of the second distortion type, and calculate a quality score of the image region.
  10. 根据权利要求6-9任一项所述的装置，其特征在于， A device according to any one of claims 6-9, wherein
    所述视觉感知特征为:自然场景统计的感知特征。The visual perception feature is: a perceptual feature of natural scene statistics.
PCT/CN2017/082348 2017-04-28 2017-04-28 Method and apparatus for evaluating quality of non-reference image WO2018195891A1 (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
PCT/CN2017/082348 WO2018195891A1 (en) 2017-04-28 2017-04-28 Method and apparatus for evaluating quality of non-reference image

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2017/082348 WO2018195891A1 (en) 2017-04-28 2017-04-28 Method and apparatus for evaluating quality of non-reference image

Publications (1)

Publication Number Publication Date
WO2018195891A1 true WO2018195891A1 (en) 2018-11-01

Family

ID=63917816

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2017/082348 WO2018195891A1 (en) 2017-04-28 2017-04-28 Method and apparatus for evaluating quality of non-reference image

Country Status (1)

Country Link
WO (1) WO2018195891A1 (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113610862A (en) * 2021-07-22 2021-11-05 东华理工大学 Screen content image quality evaluation method
CN114926461A (en) * 2022-07-19 2022-08-19 湖南工商大学 Method for evaluating quality of full-blind screen content image

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101540048A (en) * 2009-04-21 2009-09-23 北京航空航天大学 Image quality evaluating method based on support vector machine
CN101777060A (en) * 2009-12-23 2010-07-14 中国科学院自动化研究所 Automatic evaluation method and system of webpage visual quality
CN103076334A (en) * 2013-01-25 2013-05-01 上海理工大学 Method for quantitatively evaluating perceived quality of digital printed lines and texts
US20130294698A1 (en) * 2010-12-21 2013-11-07 Microsoft Corporation Image quality assessment
CN103745454A (en) * 2013-12-17 2014-04-23 北京大学 Method for evaluating image processing algorithms or systems based on rank learning
CN103778636A (en) * 2014-01-22 2014-05-07 上海交通大学 Feature construction method for non-reference image quality evaluation
CN106156777A (en) * 2015-04-23 2016-11-23 华中科技大学 Textual image detection method and device

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101540048A (en) * 2009-04-21 2009-09-23 北京航空航天大学 Image quality evaluating method based on support vector machine
CN101777060A (en) * 2009-12-23 2010-07-14 中国科学院自动化研究所 Automatic evaluation method and system of webpage visual quality
US20130294698A1 (en) * 2010-12-21 2013-11-07 Microsoft Corporation Image quality assessment
CN103076334A (en) * 2013-01-25 2013-05-01 上海理工大学 Method for quantitatively evaluating perceived quality of digital printed lines and texts
CN103745454A (en) * 2013-12-17 2014-04-23 北京大学 Method for evaluating image processing algorithms or systems based on rank learning
CN103778636A (en) * 2014-01-22 2014-05-07 上海交通大学 Feature construction method for non-reference image quality evaluation
CN106156777A (en) * 2015-04-23 2016-11-23 华中科技大学 Textual image detection method and device

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113610862A (en) * 2021-07-22 2021-11-05 东华理工大学 Screen content image quality evaluation method
CN113610862B (en) * 2021-07-22 2023-08-01 东华理工大学 Screen content image quality assessment method
CN114926461A (en) * 2022-07-19 2022-08-19 湖南工商大学 Method for evaluating quality of full-blind screen content image

Similar Documents

Publication Publication Date Title
Korshunov et al. Vulnerability assessment and detection of deepfake videos
Korshunov et al. Deepfakes: a new threat to face recognition? assessment and detection
WO2020038136A1 (en) Facial recognition method and apparatus, electronic device and computer-readable medium
CN107123122B (en) No-reference image quality evaluation method and device
KR20210134528A (en) Video processing method, apparatus, electronic device and storage medium and computer program
CN102014295B (en) Network sensitive video detection method
CN110267061B (en) News splitting method and system
CN104657468B (en) The rapid classification method of video based on image and text
WO2020124701A1 (en) Temperature measurement method and apparatus
WO2014194620A1 (en) Method, module, apparatus and system for extracting, training and detecting image characteristic
WO2021151313A1 (en) Method and apparatus for document forgery detection, electronic device, and storage medium
WO2021217899A1 (en) Method, apparatus, and device for encrypting display information, and storage medium
WO2021175040A1 (en) Video processing method and related device
KR20110080224A (en) Obscene contents filtering apparatus and method of the same
CN104661037A (en) Tampering detection method and system for compressed image quantization table
WO2024091080A1 (en) Automatic video generation method and automatic video generation server
WO2018195891A1 (en) Method and apparatus for evaluating quality of non-reference image
WO2015126097A1 (en) Interactive server and method for controlling the server
CN112990031A (en) Method for detecting tampered face video and image based on improved twin network
WO2016076515A1 (en) Method and device for generating metadata including frequency characteristic information of image
Xiao et al. Securing the socio-cyber world: Multiorder attribute node association classification for manipulated media
WO2023123923A1 (en) Human body weight identification method, human body weight identification device, computer device, and medium
Huang et al. A novel video forgery detection model based on triangular polarity feature classification
Fernández et al. Digital video manipulation detection technique based on compression algorithms
CN109040784A (en) Commercial detection method and device

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 17906799

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 14.02.2020)

122 Ep: pct application non-entry in european phase

Ref document number: 17906799

Country of ref document: EP

Kind code of ref document: A1