WO2016011589A1 - 图像分类方法和图像分类装置 - Google Patents

图像分类方法和图像分类装置 Download PDF

Info

Publication number
WO2016011589A1
WO2016011589A1 PCT/CN2014/082657 CN2014082657W WO2016011589A1 WO 2016011589 A1 WO2016011589 A1 WO 2016011589A1 CN 2014082657 W CN2014082657 W CN 2014082657W WO 2016011589 A1 WO2016011589 A1 WO 2016011589A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
images
quality
quality category
category
Prior art date
Application number
PCT/CN2014/082657
Other languages
English (en)
French (fr)
Inventor
李长宁
张焰焰
慕岳衷
Original Assignee
宇龙计算机通信科技(深圳)有限公司
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 宇龙计算机通信科技(深圳)有限公司 filed Critical 宇龙计算机通信科技(深圳)有限公司
Priority to CN201480080711.2A priority Critical patent/CN106575223B/zh
Priority to EP14898089.9A priority patent/EP3196758B1/en
Priority to PCT/CN2014/082657 priority patent/WO2016011589A1/zh
Priority to US15/327,626 priority patent/US10289939B2/en
Publication of WO2016011589A1 publication Critical patent/WO2016011589A1/zh

Links

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00Image analysis
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00Pattern recognition
    • G06F18/20Analysing
    • G06F18/22Matching criteria, e.g. proximity measures
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00Pattern recognition
    • G06F18/20Analysing
    • G06F18/24Classification techniques
    • G06F18/241Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2413Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00Pattern recognition
    • G06F18/20Analysing
    • G06F18/24Classification techniques
    • G06F18/243Classification techniques relating to the number of classes
    • G06F18/2431Multiple classes
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F9/00Arrangements for program control, e.g. control units
    • G06F9/06Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
    • G06F9/44Arrangements for executing specific programs
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00Image analysis
    • G06T7/0002Inspection of images, e.g. flaw detection
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00Arrangements for image or video recognition or understanding
    • G06V10/98Detection or correction of errors, e.g. by rescanning the pattern or by human intervention; Evaluation of the quality of the acquired patterns
    • G06V10/993Evaluation of the quality of the acquired pattern
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00Indexing scheme for image analysis or image enhancement
    • G06T2207/10Image acquisition modality
    • G06T2207/10024Color image
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00Indexing scheme for image analysis or image enhancement
    • G06T2207/20Special algorithmic details
    • G06T2207/20021Dividing image into blocks, subimages or windows
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00Indexing scheme for image analysis or image enhancement
    • G06T2207/20Special algorithmic details
    • G06T2207/20048Transform domain processing
    • G06T2207/20056Discrete and fast Fourier transform, [DFT, FFT]
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00Indexing scheme for image analysis or image enhancement
    • G06T2207/20Special algorithmic details
    • G06T2207/20076Probabilistic image processing
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00Indexing scheme for image analysis or image enhancement
    • G06T2207/30Subject of image; Context of image processing
    • G06T2207/30168Image quality inspection
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00Indexing scheme for image analysis or image enhancement
    • G06T2207/30Subject of image; Context of image processing
    • G06T2207/30196Human being; Person
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00Indexing scheme for image analysis or image enhancement
    • G06T2207/30Subject of image; Context of image processing
    • G06T2207/30196Human being; Person
    • G06T2207/30201Face

Definitions

  • the present invention relates to the field of image processing technologies, and in particular to an image classification method and an image classification device. Background technique
  • the generally defined low quality includes image blurring due to jitter or out-of-focus, relative image movement, and image smearing. Too much noise in the night scene will affect the perception of the user in terms of clarity.
  • the present invention is based on the above problems, and proposes a new technical solution, which can classify images according to the quality of the image.
  • the present invention provides an image classification method, including: an analysis step of analyzing the quality of any image; a determination step of determining a quality category to which any of the images belongs according to the analysis result, wherein the quality The category includes the first quality category suggested to be saved and the second quality category suggested to be deleted.
  • the image can be judged according to the characteristics of the image, such as sharpness. Quality, classify images by image quality, and recommend saving image categories that meet quality requirements. It is recommended to delete image categories that do not meet quality requirements. In this way, the image can be classified, and the entire category can be saved or deleted, which saves the user's operation time, improves the convenience of the user operation, and deletes the image category that does not meet the quality requirement, thereby saving a large amount of storage space.
  • the analyzing step specifically includes: calculating a resolution of the any image according to a preset definition calculation function; the determining step specifically includes: the resolution of the any image is less than Determining any of the images belonging to the second quality category when the image clarity is preset; determining any of the images when the resolution of the any image is greater than or equal to the preset image sharpness Belongs to the first quality category.
  • the sharpness of an image can be used as a quality standard of an image. Normally, a clear image contains more information than an unclear image.
  • Using the evaluation function as a preset definition calculation function can reflect a value that characterizes the sharpness of the image. The reference image when the evaluation function is at the maximum value is The best sharpness image.
  • method functions for evaluating sharpness include gradient functions, spectral functions, and entropy functions.
  • the Tenengrad function uses the soble operator to extract the gradient values of the edges.
  • the sum of the squares of the gradient values is:
  • G x, and (x'_y) are the approximations of the gradient of the image in the horizontal and vertical directions, respectively.
  • the spectral function can be used to analyze the spatial frequency of the image using a two-dimensional Fourier transform, so that the high-frequency component and the low-frequency component of the image are separated, and the clear image will be sharper and the jump will change sharply.
  • the edge contains more high-frequency components, so you can determine the sharpness of the image by evaluating the high-frequency components of the image.
  • the calculation of the resolution using the spectral function is very large and may affect the efficiency of image processing.
  • the method further includes: acquiring all images, according to all The order of the images sequentially determines the similarity between the current image and the previous image of the current image.
  • the similarity is greater than or equal to the preset similarity
  • the current image is defined as the same as the previous image.
  • a group image wherein the current image and the previous image are defined as different group images when the similarity is less than the preset similarity.
  • determining the similarity between the current image and the previous image of the current image specifically: dividing the current image and the previous image into a plurality of macroblocks, respectively The RGB values of the macroblocks of each image are converted into YUV luminance and chrominance components; the mean squared difference of the YUV components of the corresponding macroblocks in the current image and the previous image are respectively calculated, and all macros are The YUV mean square values of the blocks are summed to obtain the similarity between the current image and the previous image.
  • the image may be divided into a plurality of 32x32 macroblocks, the RGB values of the picture macroblock are obtained and converted into YUV luminance and chrominance components, and the mean square error of the YUV components of the corresponding macroblocks in the two pictures are respectively calculated. The value, and the YUV mean squared difference of all macroblocks is added. When the summation value is less than a certain threshold, the two pictures are considered to be similar.
  • the method further includes: counting the number of images in each of the same group of images; determining the number of images in any of the same group when the number of images in any of the same group of images is one The image belongs to the first quality category, or prompts the user to select a quality category to which the image in any of the same group of images belongs, and determines, according to the user's selection, that the image in any of the same group of images belongs to the a first quality category or the second quality category; when the number of images in any of the same group of images is plural, comparing the quality of the plurality of images, and ranking the plurality of images according to the comparison result Determining a predetermined number of images in the plurality of images as belonging to the first quality category, and determining other images in the plurality of images as belonging to the The second quality category is described.
  • the user may wish to store it, and therefore, when the number of images in any of the same group of images is detected as one Zhang Shi, can directly into the first quality category, save the group recommendations, can also prompt the user that the quality of the group of images is low, should be classified into the second quality category, and provide users with the classification
  • the first quality category and the second quality category are two options, and the user determines the classification of the group of images according to their own needs. When a user shoots multiple images of the same content but different quality for a certain scene, they are generally classified into the same group.
  • the images in the image grouping can be sorted according to the image quality, and a preset number is set, and the ranking is advanced. A preset number of images of relatively high quality are classified into the first quality category, and the remaining images in the image group that are ranked lower, that is, the relatively lower quality, are classified into the second quality category. In this way, it is guaranteed that images that may be useful to the user are not lost, and that it saves storage space by storing images with too much content.
  • the quality of the plurality of images is compared, and the plurality of images are ranked according to the comparison result, specifically including : determining whether a main character of the plurality of images is a person when the number of images in the same group of images is a plurality of images; and determining a result that the protagonist of the plurality of images is a person, each of the plurality of images
  • the human body image information of the image is detected to obtain a first detection result, wherein the human body image information includes an overall area and position information of the image occupied by the human body, face information, and facial features information; according to the first detection result,
  • the plurality of images are ranked.
  • the human body detection is first performed, and it is determined whether the main character and the position of the image occupied by the human body are the protagonists in the image, and then the face detection and the facial features are detected on the main character, if the detection Less than the face and facial features, indicating that the photo is not the photo that the user wants, it is probably a photo taken by mistake, it can be placed in the low quality photo of the second quality category, and for the face and facial features Normally detected photos can be further evaluated by dimensional detection such as exposure detection, sharpness detection, and noise detection to determine the quality of the photos, thereby ranking the images within the group.
  • the exposure detection determines whether the image is overexposed by determining whether the brightness average value of the image is greater than the threshold value, and the overexposed photo can be deleted by the exposure detection in the second quality category.
  • the average brightness is equal to the sum of the total value of the photo brightness/the total number of pixels and the scene exposure factor.
  • the scene exposure coefficient is related to the specific scene. Considering the different brightness values of different scenes, the concept of scene exposure coefficient is designed. For example, when the scene is snow and sunlight, the brightness is high, and the brightness of the scene itself is higher. The smaller the coefficient, the greater the average brightness obtained in scenes such as snow and sunlight.
  • Sharpness detection is to evaluate the image quality by the edge sharpness detection algorithm and the gray level change of a certain edge direction of the statistical image. That is, the sharper the gray level change, the sharper the edge and the higher the image quality.
  • Noise detection refers to the recognition of noise based on the gray correlation coefficient between the noisy image and the corresponding mean image at each pixel.
  • the method further includes: When the judgment result is that the main character of the plurality of images is an object, the brightness information, the sharpness information, and the noise information of each of the plurality of images are detected to obtain a second detection result; according to the second The detection result is to rank the plurality of images.
  • the main characters of the plurality of images are objects
  • dimension detection such as exposure detection, sharpness detection, and noise detection is performed on the plurality of images to determine the quality of the photos, thereby ranking the images in the group.
  • the method further includes: displaying, according to the received display command, the image in the first quality category and the image in the second quality category.
  • the image in the first quality category and the image in the second quality category are displayed separately, so that the user can separately operate the images of the two categories.
  • the method further includes: changing, according to the received image category change command, the quality category to which the any image belongs to the first quality category to the second quality category, or The second quality category is changed to the first quality category.
  • the system discriminates and classifies the image based on logic and algorithm. After the system completes the classification of the image, the user can still change the classification of the image according to his own needs. For example, the user temporarily decides to retain the second quality. An image in the category is no longer deleted, and the image can be recalled from the second quality category to the first quality category. The image is saved.
  • the method further includes: deleting all the images in the second quality category according to the received deletion command.
  • the image in the second quality category is uniformly deleted, which greatly improves the efficiency of deleting the image, so that the user does not need to judge the image quality one by one and then delete the image.
  • an image classification apparatus including: an analysis unit that analyzes a quality of any image; and a determination unit that determines, according to the analysis result, a quality category to which the image belongs, where The quality category includes a first quality category suggested to be saved and a second quality category suggested to be deleted.
  • the quality of the image can be judged according to the characteristics of the image (such as sharpness), the image is classified by the quality of the image, and it is recommended to save the image category that meets the quality requirement, and it is recommended to delete the image category that does not meet the quality requirement. .
  • the characteristics of the image such as sharpness
  • the image is classified by the quality of the image, and it is recommended to save the image category that meets the quality requirement, and it is recommended to delete the image category that does not meet the quality requirement.
  • the analyzing unit includes: a calculating unit, calculating a sharpness of the any image according to a preset sharpness calculating function; and the determining unit is configured to: When the resolution is less than the preset image definition, determining that any of the images belongs to the second quality category, and determining that the resolution of the image is greater than or equal to the preset image resolution Any image belongs to the first quality category.
  • the sharpness of an image can be used as a quality standard of an image. Normally, a clear image contains more information than an unclear image.
  • Using the evaluation function as a preset definition calculation function can reflect a value that characterizes the sharpness of the image.
  • the reference image at the evaluation function at the maximum is clear. The best image.
  • method functions for evaluating sharpness include gradient functions, spectral functions, and entropy functions.
  • the Tenengrad function uses the soble operator to extract the gradient values of the edges.
  • the sum of the squares of the gradient values is:
  • G x, and (x'_y) are the approximations of the gradient of the image in the horizontal and vertical directions, respectively.
  • the spectral function can be used to analyze the spatial frequency of the image using a two-dimensional Fourier transform, so that the high-frequency component and the low-frequency component of the image are separated, and the clear image will be sharper and the jump will change sharply.
  • the edge contains more high-frequency components, so you can determine the sharpness of the image by evaluating the high-frequency components of the image.
  • the calculation of the resolution using the spectral function is very large and may affect the efficiency of image processing.
  • the method further includes: a determining unit, acquiring all images, sequentially determining, according to the order of all the images, a similarity between the current image and the previous image of the current image;
  • a determining unit acquiring all images, sequentially determining, according to the order of all the images, a similarity between the current image and the previous image of the current image;
  • the similarity is greater than or equal to the preset similarity
  • the current image and the previous image are defined as the same group of images
  • the similarity is less than the preset similarity
  • the current image is The previous image is defined as a different set of images.
  • the determining unit includes: a dividing unit, respectively dividing the current image and the previous image into a plurality of macroblocks, acquiring RGB values of macroblocks of each image, and converting a YUV luminance and chrominance component; a similarity calculation unit, respectively calculating a mean square difference of the YUV components of the corresponding macroblock in the current image and the previous image, and calculating a YUV mean square error of all the macroblocks The values are summed to obtain the similarity between the current image and the previous image.
  • the image may be divided into a plurality of 32x32 macroblocks, the RGB values of the picture macroblock are obtained and converted into YUV luminance and chrominance components, and the mean square error of the YUV components of the corresponding macroblocks in the two pictures are respectively calculated. The value, and the YUV mean squared difference of all macroblocks is added. When the summation value is less than a certain threshold, the two pictures are considered to be similar.
  • the method further includes: a statistical unit, counting the number of images in each of the same group of images; and the determining unit is further configured to: when the number of images in any of the same group of images is one And determining that the image in the image of any of the same group belongs to the first quality category, or prompting the user to select a quality category to which the image in the same group of images belongs, and determining the location according to the selection of the user
  • the image classification device further includes: a quality comparison unit, when the number of images in any of the same group of images is And comparing the quality of the plurality of images, and ranking the plurality of images according to the comparison result; the determining unit is further configured to: determine a preset number of images ranked in the plurality of the plurality of images In order to belong to the first quality category, other images in the plurality of images are determined to belong to the second quality category.
  • the user may wish to store it, and therefore, when the number of images in any of the same group of images is detected as one Zhang Shi, can directly into the first quality category, save the group recommendations, can also prompt the user that the quality of the group of images is low, should be classified into the second quality category, and provide users with the classification
  • the first quality category and the second quality category are two options, and the user determines the classification of the group of images according to their own needs. When a user shoots multiple images of the same content but different quality for a certain scene, they are generally classified into the same group.
  • the images in the image grouping can be sorted according to the image quality, and a preset number is set, and the ranking is advanced. A preset number of images of relatively high quality are classified into the first quality category, and the remaining images in the image group that are ranked lower, that is, the relatively lower quality, are classified into the second quality category. In this way, it is guaranteed that images that may be useful to the user are not lost, and that it saves storage space by storing images with too much content.
  • the quality comparison unit includes: a main angle determining unit, when the number of images in the same group of images is multiple, determining whether the main character of the plurality of images is a person; a unit, when the result of the determination is that the main character of the plurality of images is a human, detecting human body image information of each of the plurality of images to obtain a first detection result, wherein the human body image information includes a human body occupation Overall area and location letter of the image Information, facial information and facial features information; a first ranking unit, ranking the plurality of images according to the first detection result.
  • the human body detection is first performed, and it is determined whether the main character and the position of the image occupied by the human body are the protagonists in the image, and then the face detection and the facial features are detected on the main character, if the detection Less than the face and facial features, indicating that the photo is not the photo that the user wants, it is probably a photo taken by mistake, it can be placed in the low quality photo of the second quality category, and for the face and facial features Normally detected photos can be further evaluated by dimensional detection such as exposure detection, sharpness detection, and noise detection to determine the quality of the photos, thereby ranking the images within the group.
  • the quality comparison unit further includes: a second detecting unit, when the judgment result is that the main character of the plurality of images is an object, brightness information of each image in the plurality of images And detecting the sharpness information and the noise information to obtain the second detection result; and the second ranking unit, ranking the plurality of images according to the second detection result.
  • the main characters of the plurality of images are objects
  • dimension detection such as exposure detection, sharpness detection, and noise detection is performed on the plurality of images to determine the quality of the photos, thereby ranking the images in the group.
  • the method further includes: displaying, by the display unit, the image in the first quality category and the image in the second quality category are displayed in a differentiated manner according to the received display command.
  • the image in the first quality category and the image in the second quality category are displayed separately, so that the user can separately operate the images of the two categories.
  • the method further includes: a category changing unit, changing, according to the received image category change command, the quality category to which the any image belongs from the first quality category to the second quality category Or changing from the second quality category to the first quality category.
  • the system discriminates and classifies the image based on logic and algorithm. After the system completes the classification of the image, the user can still change the classification of the image according to his own needs. For example, the user temporarily decides to retain the second quality. An image in the category is no longer deleted, and the image can be recalled from the second quality category to the first quality category. The image is saved.
  • the method further includes: deleting the unit, and deleting all the images in the second quality category according to the received deletion command.
  • the image in the second quality category is uniformly deleted, which greatly improves the efficiency of deleting the image, so that the user does not need to judge the image quality one by one and then delete the image.
  • the image can be classified according to the quality of the image, and the image that is needed by the user is distinguished from the image that is not needed, and the image that the user does not need is uniformly deleted, so that the operation of the user is simple and convenient, and the storage space can be saved.
  • Figure 1 shows a prior art mobile phone album interface.
  • FIG. 2 shows a flow chart of an image classification method according to an embodiment of the present invention
  • FIG. 3 shows a block diagram of an image classification device in accordance with an embodiment of the present invention
  • FIG. 4 is a schematic diagram showing a contrast between a sharp image and a blurred image according to an embodiment of the present invention
  • 5A and 5B illustrate a mobile phone album interface in accordance with an embodiment of the present invention. detailed description
  • FIG. 2 shows a flow chart of an image classification method according to an embodiment of the present invention.
  • an image classification method includes:
  • Step 202 analyzing quality of any image
  • Step 204 Determine, according to the analysis result, a quality category to which any image belongs, where the quality category includes a first quality category suggested to be saved and a second quality category suggested to be deleted.
  • the quality of the image can be judged according to the characteristics of the image (such as sharpness), the image is classified by the quality of the image, and it is recommended to save the image category that meets the quality requirement, and it is recommended to delete the image category that does not meet the quality requirement. .
  • the image can be classified, and the entire category can be saved or deleted, which saves the user's operation time and improves the convenience of the user operation. Deleting the image category that does not meet the quality requirement can save a lot of storage space.
  • the step 202 includes: calculating a resolution of any image according to a preset definition calculation function; and the step 204 specifically includes: determining, when the resolution of any image is smaller than the preset image resolution, Any image belongs to the second quality category; when any of the images has a sharpness greater than or equal to the preset image sharpness, it is determined that any of the images belong to the first quality category.
  • the sharpness of an image can be used as a quality standard of an image. Normally, a clear image contains more information than an unclear image.
  • Using the evaluation function as a preset definition calculation function can reflect a value that characterizes the sharpness of the image. The reference image when the evaluation function is at the maximum value is The best sharpness image.
  • method functions for evaluating sharpness include gradient functions, spectral functions, and entropy functions.
  • the Tenengrad function uses the soble operator to extract the gradient values of the edges.
  • the sum of the squares of the gradient values is:
  • G x (x, and (x'_y) are the gradient approximations of the image in the horizontal and vertical directions, respectively.
  • the spectral function can be used to analyze the spatial frequency of the image using a two-dimensional Fourier transform. In this way, the high-frequency component of the image is separated from the low-frequency component, and the sharp image has sharper edges with strong jumps and more high-frequency components, so the image can be determined by evaluating the high-frequency components of the image.
  • the use of spectral functions to evaluate sharpness is computationally intensive and may affect the efficiency of image processing.
  • the method further includes: acquiring all the images, sequentially determining the similarity between the current image and the previous image of the current image according to the order of all the images, and when the similarity is greater than or equal to the preset similarity,
  • the current image and the previous image are defined as the same group of images, and when the similarity is less than the preset similarity, the current image and the previous image are defined as different sets of images.
  • determining the similarity between the current image and the previous image of the current image comprises: dividing the current image and the previous image into a plurality of macroblocks, respectively, and acquiring macroblocks of each image.
  • the RGB values are converted into YUV luminance and chrominance components; the mean squared difference of the YUV components of the corresponding macroblock in the current image and the previous image are calculated separately, and the YUV mean squared differences of all macroblocks are summed , to get the similarity between the current image and the previous image.
  • the image may be divided into a plurality of 32x32 macroblocks, the RGB values of the picture macroblock are obtained and converted into YUV luminance and chrominance components, and the mean square error of the YUV components of the corresponding macroblocks in the two pictures are respectively calculated. The value, and the YUV mean squared difference of all macroblocks is added. When the summation value is less than a certain threshold, the two pictures are considered to be similar.
  • the method further includes: counting the number of images in each of the same group of images; and step 204 further comprising: determining any of the same group of images when the number of images in any of the same group of images is one The image in the image belongs to the first quality category, or prompts the user to select the quality category to which the image in any of the same group of images belongs, and determines that the image in any of the same group of images belongs to the first quality category or the second quality according to the user's selection.
  • the quality of the plurality of images is compared, and the plurality of images are ranked according to the comparison result; the preset number of the plurality of images is ranked first
  • the image is determined to belong to the first quality category, and the other images in the plurality of images are determined to belong to the second quality category.
  • the images in the image grouping can be sorted according to the image quality, and a preset number is set, and the ranking is advanced. A preset number of images of relatively high quality are classified into the first quality category, and the remaining images in the image group that are ranked lower, that is, the relatively lower quality, are classified into the second quality category. In this way, it is ensured that the image that may be useful to the user is not lost, and that the storage of too much content is avoided, and the storage space is wasted.
  • the quality of the plurality of images is compared, and the plurality of images are ranked according to the comparison result, specifically: when the same group When the number of images in the image is multiple, it is determined whether the main character of the plurality of images is a person; when the judgment result is that the main character of the plurality of images is a person, the human body image information of each image in the plurality of images is detected to obtain the first a detection result, wherein the human body image information includes an overall area and position information of the image occupied by the human body, face information, and facial features information; and the plurality of images are ranked according to the first detection result.
  • the human body detection is first performed, and it is determined whether the main character and the position of the image occupied by the human body are the protagonists in the image, and then the face detection and the facial features are detected on the main character, if the detection Less than the face and facial features, indicating that the photo is not the photo that the user wants, it is probably a photo taken by mistake, it can be placed in the low quality photo of the second quality category, and for the face and facial features Normally detected photos can be further evaluated by dimensional detection such as exposure detection, sharpness detection, and noise detection to determine the quality of the photos, thereby ranking the images within the group.
  • the exposure detection determines whether the image is overexposed by determining whether the brightness average value of the image is greater than the threshold value, and the overexposed photo can be placed in the second quality category by exposure detection.
  • the average brightness is equal to the sum of the total value of the photo brightness/the total number of pixels and the scene exposure factor.
  • the scene exposure coefficient is related to the specific scene. Considering the different brightness values of different scenes, the concept of scene exposure coefficient is designed. For example, when the scene is snow and sunlight, the brightness is high, and the brightness of the scene itself is higher. The smaller the coefficient, the greater the average brightness obtained in scenes such as snow and sunlight.
  • Sharpness detection is to evaluate the image quality by the edge sharpness detection algorithm and the gray level change of a certain edge direction of the statistical image. That is, the sharper the gray level change, the sharper the edge and the higher the image quality.
  • Noise detection refers to the recognition of noise based on the gray correlation coefficient between the noisy image and the corresponding mean image at each pixel.
  • the method further includes: when the judgment result is When the main character of the plurality of images is an object, the brightness information, the sharpness information, and the noise information of each of the plurality of images are detected to obtain a second detection result; and the plurality of images are ranked according to the second detection result.
  • the main characters of the plurality of images are objects
  • dimension detection such as exposure detection, sharpness detection, and noise detection is performed on the plurality of images to determine the quality of the photos, thereby ranking the images in the group.
  • the method further includes: displaying, according to the received display command, the image in the first quality category and the image in the second quality category.
  • the image in the first quality category and the image in the second quality category are displayed separately, so that the user can separately operate the images of the two categories.
  • the method further includes: changing a quality category to which any image belongs from the first quality category to the second quality category according to the received image category change command, or changing from the second quality category to the second quality category A quality category.
  • the system discriminates and classifies the image based on logic and algorithm. After the system completes the classification of the image, the user can still change the classification of the image according to his own needs. For example, the user temporarily decides to retain the second quality. An image in the category is no longer deleted, and the image can be recalled from the second quality category to the first quality category, and the image is saved.
  • the method further includes: deleting according to the received delete command All images in the second quality category.
  • the image in the second quality category is uniformly deleted, which greatly improves the efficiency of deleting the image, so that the user does not need to judge the image quality one by one and then delete the image.
  • FIG. 3 shows a block diagram of an image classification device in accordance with an embodiment of the present invention.
  • the image classification device 300 includes: an analysis unit 302, which analyzes the quality of any image; and a determination unit 304, according to the analysis result, determines a quality category to which any image belongs, where The quality category includes the first quality category recommended for preservation and the second quality category suggested for deletion.
  • the quality of the image can be judged according to the characteristics of the image (such as sharpness), the image is classified by the quality of the image, and it is recommended to save the image category that meets the quality requirement, and it is recommended to delete the image category that does not meet the quality requirement. .
  • the image can be classified and the entire category can be saved or deleted, which saves the user's operation time and improves the convenience of the user operation. Deleting the image category that does not meet the quality requirement can save a lot of storage space.
  • the analyzing unit 302 includes: a calculating unit 3022, calculating a sharpness of any image according to a preset definition calculating function; and determining unit 304 for: the resolution of any image is smaller than a preset When the image is sharp, it is determined that any of the images belongs to the second quality category, and when the sharpness of any of the images is greater than or equal to the preset image sharpness, it is determined that any of the images belongs to the first quality category.
  • the sharpness of an image can be used as a quality standard of an image. Normally, a clear image contains more information than an unclear image.
  • Using the evaluation function as a preset definition calculation function can reflect a value that characterizes the sharpness of the image. The reference image when the evaluation function is at the maximum value is The best sharpness image.
  • method functions for evaluating sharpness include gradient functions, spectral functions, and entropy functions.
  • the Tenengrad function uses the soble operator to extract the gradient values of the edges.
  • the sum of the squares of the gradient values is:
  • G x, and (x'_y) are the approximations of the gradient of the image in the horizontal and vertical directions, respectively.
  • the spectral function can be used to analyze the spatial frequency of the image using a two-dimensional Fourier transform, so that the high-frequency component and the low-frequency component of the image are separated, and the clear image will be sharper and the jump will change sharply.
  • the edge contains more high-frequency components, so you can determine the sharpness of the image by evaluating the high-frequency components of the image.
  • the calculation of the resolution using the spectral function is very large and may affect the efficiency of image processing.
  • the method further includes: the determining unit 306, acquiring all the images, sequentially determining the similarity between the current image and the previous image of the current image according to the order of all the images; the grouping unit 308, the similarity is greater than or When the similarity is equal to the preset similarity, the current image and the previous image are defined as the same group image, and when the similarity is less than the preset similarity, the current image and the previous image are defined as different groups of images.
  • the determining unit 306 includes: a dividing unit 3062, respectively dividing the current image and the previous image into a plurality of macroblocks, acquiring RGB values of macroblocks of each image, and converting into YUV brightness and The chrominance component; the similarity calculation unit 3064, respectively calculating the mean square difference of the YUV component of the corresponding macroblock in the current image and the previous image, and summing the YUV mean square differences of all the macroblocks to obtain The similarity between the current image and the previous image.
  • the image may be divided into a plurality of 32x32 macroblocks, the RGB values of the picture macroblock are obtained and converted into YUV luminance and chrominance components, and the mean square error of the YUV components of the corresponding macroblocks in the two pictures are respectively calculated. The value, and the YUV mean squared difference of all macroblocks is added. When the summation value is less than a certain threshold, the two pictures are considered to be similar.
  • the method further includes: a statistical unit 310, counting the number of images in each of the same group of images; and the determining unit 304 is further configured to: when the number of images in any of the same group of images is one Determining that the image in any of the same group of images belongs to the first quality category, or prompting the user to select a quality category to which the image in any of the same group of images belongs, and determining that the image in any of the same group of images belongs to the first according to the user's selection.
  • the quality classification or the second quality category; the image classification device 300 further includes: a quality comparison unit 312, when the number of images in any of the same group of images is plural, compares the quality of the plurality of images, and compares the plurality of images according to the comparison result The image is ranked; the determining unit 304 is further configured to: determine a predetermined number of images in the plurality of images as belonging to the first quality category, and determine other images in the plurality of images as belonging to the second quality category.
  • the user may wish to store it, and therefore, when the number of images in any of the same group of images is detected as one Zhang Shi, can directly into the first quality category, save the group recommendations, can also prompt the user that the quality of the group of images is low, should be classified into the second quality category, and provide users with the classification
  • the first quality category and the second quality category are two options, and the user determines the classification of the group of images according to their own needs. When a user shoots multiple images of the same content but different quality for a certain scene, they are generally classified into the same group.
  • the images in the image grouping can be sorted according to the image quality, and a preset number is set, and the ranking is advanced. A preset number of images of relatively high quality are classified into the first quality category, and the remaining images in the image group that are ranked lower, that is, the relatively lower quality, are classified into the second quality category. In this way, it is guaranteed that images that may be useful to the user are not lost, and that it saves storage space by storing images with too much content.
  • the quality comparison unit 312 includes: a main angle determining unit 3122, when the number of images in the same group of images is plural, determining whether the main character of the plurality of images is a human; the first detecting unit 3124, When the judgment result is that the main character of the plurality of images is a person, the human body image information of each image in the plurality of images is detected to obtain a first detection result, wherein the human body image information includes an overall area and position information of the image occupied by the human body, The face information and the facial features information; the first ranking unit, ranking the plurality of images according to the first detection result.
  • the human body detection is first performed, and it is determined whether the main character and the position of the image occupied by the human body are the protagonists in the image, and then the face detection and the facial features are detected on the main character, if the detection Less than the face and facial features, indicating that the photo is not the photo that the user wants, it is probably a photo taken by mistake, it can be placed in the low quality photo of the second quality category, and for the face and facial features Normally detected photos can be further evaluated by dimensional detection such as exposure detection, sharpness detection, and noise detection to determine the quality of the photos, thereby ranking the images within the group.
  • the quality comparison unit 312 further includes: a second detecting unit 3126, when the judgment result is that the main character of the plurality of images is the object, the brightness information and the sharpness information of each image in the plurality of images And the noise information is detected to obtain a second detection result; and the second ranking unit ranks the plurality of images according to the second detection result.
  • the main characters of the plurality of images are objects
  • dimension detection such as exposure detection, sharpness detection, and noise detection is performed on the plurality of images to determine the quality of the photos, thereby ranking the images in the group.
  • the method further includes: a display unit 314, configured to distinguish the image in the first quality category and the image in the second quality category according to the received display command.
  • the image in the first quality category and the image in the second quality category are displayed separately, so that the user can separately operate the images of the two categories.
  • the method further includes: a category changing unit 316, changing a quality category to which any image belongs from the first quality category to the second quality category according to the received image category change command, or by the second The quality category is changed to the first quality category.
  • the system discriminates and classifies the image based on logic and algorithm. After the system completes the classification of the image, the user can still change the classification of the image according to his own needs. For example, the user temporarily decides to retain the second quality. An image in the category is no longer deleted, and the image can be recalled from the second quality category to the first quality category, and the image is saved.
  • the method further includes: a deleting unit 318, deleting all the images in the second quality category according to the received deletion command.
  • the image in the second quality category is uniformly deleted, which greatly improves the efficiency of deleting the image, so that the user does not need to judge the image quality one by one and then delete the image.
  • Fig. 4 shows a comparison of a clear image and a blurred image in accordance with an embodiment of the present invention.
  • a clear image contains more information than an unclear image
  • the evaluation function can be used as a preset definition calculation function to reflect a value representing the sharpness of the image.
  • the reference image at the time when the evaluation function is at the maximum value is the image with the best definition.
  • method functions for evaluating sharpness include a gradient function, a spectral function, and an entropy function.
  • the Tenengrad function uses the soble operator to extract the gradient values of the edges.
  • the sum of the squares of the gradient values is:
  • G x (x, and (x'_y) are the gradient approximations of the image in the horizontal and vertical directions, respectively.
  • the spectral function can be used to analyze the spatial frequency of the image using a two-dimensional Fourier transform. In this way, the high-frequency component of the image is separated from the low-frequency component, and the sharp image has sharper edges with strong jumps and more high-frequency components, so the image can be determined by evaluating the high-frequency components of the image.
  • the use of spectral functions to evaluate sharpness is computationally intensive and may affect the efficiency of image processing.
  • the discriminating low-quality image can be extracted and classified as a "low-quality photo". In this way, deleting "low-quality photos" that do not meet the quality requirements in batches can make the user's operation easy and save a lot of storage space.
  • FIG. 5A and 5B illustrate a mobile phone album interface in accordance with an embodiment of the present invention.
  • the "All Images” option and the "Low Quality Image” option are set on the phone album interface.
  • the search process and the identification process of low-quality images can be realized in two ways. Batch processing can be performed on all the images taken, each image is separately identified in the album, and low-quality image groups are separated, or each shot can be taken. An image is identified once and the low quality image is classified into a low quality image group.
  • Such a search and authentication method can be done in the background processing of the album. After processing, you can display the "All Images” option and the "Low Quality Image” option in the phone album interface, which gives the user a choice, allowing the user to choose whether to view all images or only low quality images.
  • the low quality image interface as shown in FIG. 5B can be obtained.
  • the interface only displays the low quality image automatically recognized by the mobile phone, so that it is displayed in front of the user, and the user decides whether or not Remove some of these images.
  • the image can be classified according to the quality of the image, and the image that the user needs and does not need is distinguished, and the image that the user does not need is unified. Deletion makes the user's operation simple and convenient, and also saves storage space.
  • a program product stored on a non-transitory machine readable medium for image classification in a terminal, the program product comprising a machine for causing a computer system to perform the following steps Executable instructions: analyzing the quality of any image; determining a quality category to which the any image belongs according to the analysis result, wherein the quality category includes a first quality category suggested to be saved and a second quality category suggested to be deleted.
  • a non-volatile machine readable medium storing a program product for image classification in a terminal, the program product comprising machine executable instructions for causing a computer system to perform the following steps: The quality of any image is analyzed; the quality category to which the any image belongs is determined according to the analysis result, wherein the quality category includes a first quality category suggested to be saved and a second quality category suggested to be deleted.
  • a machine readable program the program causing a machine to perform the image classification method according to any one of the above aspects.
  • a storage medium storing a machine readable program, wherein the machine readable program causes a machine to perform the image classification method according to any one of the above aspects.
  • the terms “first” and “second” are used for the purpose of description only, and are not to be construed as indicating or implying relative importance; the term “plurality” means two or more.
  • the specific meanings of the above terms in the present invention can be understood on a case-by-case basis.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Data Mining & Analysis (AREA)
  • Quality & Reliability (AREA)
  • General Engineering & Computer Science (AREA)
  • Software Systems (AREA)
  • Artificial Intelligence (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Multimedia (AREA)
  • Human Computer Interaction (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
  • Image Analysis (AREA)
  • Processing Or Creating Images (AREA)

Abstract

提供了一种图像分类方法,包括:分析步骤,对任一图像的质量进行分析;判定步骤,根据分析结果判定所述任一图像所属的质量类别,其中,所述质量类别包括建议保存的第一质量类别和建议删除的第二质量类别。相应地,还提出了一种图像分类装置。通过该技术方案,可以根据图像的质量将图像分类处理,以提高用户体验。

Description

图像分类方法和图像分类装置
技术领域
本发明涉及图像处理技术领域, 具体而言, 涉及一种图像分类方法和 一种图像分类装置。 背景技术
用户在拍摄图像时, 常常由于拍摄设备的质量原因或用户的拍摄技术 原因, 拍摄出低质量的图像, 一般定义的低质量包括因抖动或者失焦造成 图像模糊、 相对移动造成图像有拖影以及夜景中的噪点过多, 多会从清晰 度上对用户的感知造成影响。 人们用相机、 手机或平板电脑拍完图像以 后, 可以逐个浏览拍摄到的图像, 如图 1 所示, 在手机相册界面上列出了 多张图像, 用户可以打开每张图像对其质量进行判断, 当判断出拍摄的图 像质量不好时, 可以将质量不好的图像逐个删除, 并对剩下的图像进行硬 盘备份或云端备份。 然而, 这样逐个查看并鉴别图像十分耽误时间, 有些 情况下, 用户不去鉴别图像的质量, 也就不会去删除任何图像, 甚至在没 有对图像进行查看并鉴别的情况下, 就直接对图像进行了备份, 这样就导 致备份了很多质量低的图像, 极大地浪费了存储空间。
因此, 如何对图像进行分类处理, 成为目前亟待解决的问题。 发明内容
本发明正是基于上述问题, 提出了一种新的技术方案, 可以根据图像 的质量将图像分类处理。
有鉴于此, 本发明提出了一种图像分类方法, 包括: 分析步骤, 对任 一图像的质量进行分析; 判定步骤, 根据分析结果判定所述任一图像所属 的质量类别, 其中, 所述质量类别包括建议保存的第一质量类别和建议删 除的第二质量类别。
在该技术方案中, 可以根据图像的特性 (比如清晰度) 来判断图像的 质量, 通过图像的质量对图像进行分类, 并建议保存符合质量要求的图像 类别, 建议删除不符合质量要求的图像类别。 这样, 就可以对图像进行分 类, 并对整个类别进行保存或删除操作, 节省了用户的操作时间, 提升了 用户操作的便利性, 同时删除不符合质量要求的图像类别可以节省大量的 存储空间。
在上述技术方案中, 优选地, 所述分析步骤具体包括: 根据预设清晰 度计算函数计算所述任一图像的清晰度; 所述判定步骤具体包括: 在所述 任一图像的清晰度小于预设的图像清晰度时, 判定所述任一图像属于所述 第二质量类别; 在所述任一图像的清晰度大于或等于所述预设的图像清晰 度时, 判定所述任一图像属于所述第一质量类别。
在该技术方案中, 可以将图像的清晰度作为图像的质量标准。 通常情 况下, 清晰的图像比不清晰的图像包含更多的信息, 使用评价函数作为预 设清晰度计算函数可以反映表征图像清晰度的一个数值, 在评价函数处于 最大值时的参考图像即为清晰度最佳的图像。 目前, 评价清晰度的方法函 数有梯度函数、 频谱函数以及熵函数等。
例如, 釆用梯度函数中的 Tenengrad 函数来计算图像水平和垂直方向 的边缘梯度值, 并用其平方和来设定图像的清晰度范围。 Tenengrad 函数 使用 soble算子来提取边缘的梯度值, 梯度值的平方和为:
Figure imgf000003_0001
y 其中 S ( x,y )是在点 (x,y ) 上与 Soble算子的卷积, 其计算公式为:
Figure imgf000003_0002
其中, G x, 和 (x'_y)分别为图像在横向与纵向上的梯度近似值。 另外, 还可以釆用频谱函数, 使用二维的傅里叶变换对图像的空间频 率进行分析, 以此将图像的高频分量与低频分量分开, 清晰的图像会有更 尖锐、 跳变更强烈的边缘, 包含更多的高频分量, 所以可以通过评估图像 的高频分量来决定图像的清晰度。 但是釆用频谱函数评估清晰度的计算量 很大, 可能影响图像处理的效率。
在上述技术方案中, 优选地, 还包括: 获取所有图像, 按照所述所有 图像的顺序依次判断当前图像与所述当前图像的前一张图像的相似度, 在 所述相似度大于或等于预设相似度时, 将所述当前图像与所述前一张图像 定义为同组图像, 在所述相似度小于所述预设相似度时, 将所述当前图像 与所述前一张图像定义为不同组图像。
在该技术方案中, 由于用户经常会对某一场景拍摄多张质量不同的图 像, 有必要首先判别多个图像是否是属于同一个场景, 然后再对同场景的 图像进行鉴别。 同场景的图像由于其相似度较高, 即使有质量差别, 其清 晰度差别也不会太大, 所以经过初次判断后, 这些图像一般都被归入同一 组别, 如果直接对这些拍摄内容一致的图像进行保存, 会占用大量存储空 间, 而如果直接对这些拍摄内容一致的图像进行删除, 又难保用户不会用 到这些图像。 因此, 还应当对这些内容重复的图像进行再次筛选。
在上述技术方案中, 优选地, 判断当前图像与所述当前图像的前一张 图像的相似度, 具体包括: 分别将所述当前图像和所述前一张图像划分为 多个宏块, 获取每个图像的宏块的 RGB 值并转换成 YUV 亮度和色度分 量; 分别计算所述当前图像和所述前一张图像中对应的宏块的 YUV 分量 的均方差值, 并将所有宏块的 YUV 均方差值进行加和, 以得到所述当前 图像和所述前一张图像的相似度。
在该技术方案中, 可以将图像划分为若干 32x32的宏块, 获取图片宏 块的 RGB值并转换成 YUV亮度和色度分量, 分别计算两幅图片中对应宏 块的 YUV 分量的均方差值, 并将所有宏块的 YUV 均方差值进行加和, 当加和值小于某一阔值时则认为两幅图片比较相似。
在上述技术方案中, 优选地, 还包括: 统计每个所述同组图像中图像 的数量; 当任一同组图像中图像的数量为一张时, 判定所述任一同组图像 中的图像属于所述第一质量类别, 或者提示用户选择所述任一同组图像中 的图像所属的质量类别, 并根据所述用户的选择判定所述任一同组图像中 的图像属于所述第一质量类别或所述第二质量类别; 当所述任一同组图像 中图像的数量为多张时, 将多张图像的质量进行比较, 并根据比较结果对 所述多张图像进行排名; 将所述多张图像中排名靠前的预设数量的图像判 定为属于所述第一质量类别, 将所述多张图像中的其他图像判定为属于所 述第二质量类别。
在该技术方案中, 当对某场景拍摄的图像只有一张时, 即使该图像质 量较低, 用户也可能希望将其存储起来, 因此, 当检测到任一同组图像中 图像的数量为一张时, 可以直接将该组归入第一质量类别, 对该组建议保 存, 也可以向用户提示该组图像的质量较低, 应被归入第二质量类别, 并 为用户提供归入第一质量类别和归入第二质量类别两个选项, 由用户根据 自身需要来决定该组图像的分类。 用户对某一场景拍摄多张内容相同但质 量不同的图像由于其相似度一般被归在同一组, 如果将这样的图像分组直 接保存, 不仅会占用大量空间, 而且保存大量重复内容也没有实际意义, 而如果将这样的图像分组直接删除, 又难保用户不会用到这些图像, 因 此, 可以根据图像质量对该图像分组内的图像进行排序, 设定一个预设数 量, 将排名靠前也就是质量相对较高的预设数量的图像归入第一质量类 别, 将该图像分组内其余排名靠后也就是质量相对较低的图像归入第二质 量类别。 这样, 既可以保证不会丟失对用户可能有用的图像, 也避免了存 储过多内容重复的图像而浪费存储空间。
在上述技术方案中, 优选地, 当所述任一同组图像中图像的数量为多 张时, 将多张图像的质量进行比较, 并根据比较结果对所述多张图像进行 排名, 具体包括: 当所述同一组图像中图像的数量为多张时, 判断所述多 张图像的主角是否为人; 在判断结果为所述多张图像的主角为人时, 对所 述多张图像中每张图像的人体图像信息进行检测, 以得到第一检测结果, 其中, 所述人体图像信息包括人体占据的图像的整体面积和位置信息、 人 脸信息和五官信息; 根据所述第一检测结果, 对所述多张图像进行排名。
在该技术方案中, 当主角为人时, 首先进行人体检测, 根据检测到人 体占据的图像的整体面积和位置判断是否是图像中的主角, 然后再对主角 进行人脸检测和五官检测, 若检测不到人脸和五官, 表明该照片不是用户 想要的照片, 很可能是误拍的照片, 可以把它放到第二质量类别的低质量 照片中备选, 而对于人脸和五官都可以正常检测的照片可以再运用曝光检 测、 锐度检测和噪点检测等维度检测来判断照片的质量, 从而对分组内的 图像进行排名。 其中, 曝光检测为通过判断图像的亮度平均值是否大于阔值来判断图 像是否过曝, 通过曝光检测可以将过曝的照片放入第二质量类别建议删 除。 亮度平均值等于照片亮度的总值 /像素点总数与场景曝光系数的乘 积。 场景曝光系数与具体场景有关, 考虑到不同场景本身的亮度值不同, 设计了场景曝光系数的概念, 比如场景为白雪和阳光的情况下本身亮度就 很高, 而场景本身亮度越高, 场景曝光系数就越小, 以避免在白雪和阳光 等场景下得到的亮度平均值超过阔值。 锐度检测是通过对边缘锐度的检测 算法, 统计图像的某一边缘方向的灰度变化情况来评价图像质量, 即灰度 变化越剧烈, 边缘就越清晰, 图像质量就越高。 噪点检测则是指根据含噪 图像与对应均值图像在各像素点的灰色关联系数来识别噪点。
上述技术方案中, 优选地, 当所述任一同组图像中图像的数量为多张 时, 将多张图像的质量进行比较, 并根据比较结果对所述多张图像进行排 名, 还包括: 当判断结果为所述多张图像的主角为物时, 对所述多张图像 中每张图像的亮度信息、 锐度信息以及噪点信息进行检测, 以得到第二检 测结果; 根据所述第二检测结果, 对所述多张图像进行排名。
在该技术方案中, 当多张图像的主角为物时, 对多张图像进行曝光检 测、 锐度检测和噪点检测等维度检测来判断照片的质量, 从而对分组内的 图像进行排名。
在上述技术方案中, 优选地, 还包括: 根据接收到的显示命令, 对所 述第一质量类别中的图像和所述第二质量类别中的图像进行区分显示。
在该技术方案中, 将第一质量类别中的图像和第二质量类别中的图像 区分显示, 便于用户对两个类别的图像分别进行操作。
在上述技术方案中, 优选地, 还包括: 根据接收到的图像类别更改命 令, 将所述任一图像所属的质量类别由所述第一质量类别更改为所述第二 质量类别, 或者由所述第二质量类别更改为所述第一质量类别。
在该技术方案中, ***对图像判别和分类是完全依据逻辑和算法的, 在***完成对图像的分类后, 用户仍可以根据自身的需求改变图像的分 类, 比如, 用户临时决定保留第二质量类别中的某一图像, 不再对它进行 删除操作, 此时便可以将该图像从第二质量类别调回第一质量类别, 将该 图像保存。
在上述技术方案中, 优选地, 还包括: 根据接收到的删除命令, 删除 所述第二质量类别中的所有图像。
在该技术方案中, 对第二质量类别中的图像进行统一删除, 大大提高 了删除图像的效率, 使用户不用再逐个判断图像质量再删除图像。
根据本发明的另一方面, 还提供了一种图像分类装置, 包括: 分析单 元, 对任一图像的质量进行分析; 判定单元, 根据分析结果判定所述任一 图像所属的质量类别, 其中, 所述质量类别包括建议保存的第一质量类别 和建议删除的第二质量类别。
在该技术方案中, 可以根据图像的特性 (比如清晰度) 来判断图像的 质量, 通过图像的质量对图像进行分类, 并建议保存符合质量要求的图像 类别, 建议删除不符合质量要求的图像类别。 这样, 通过对图像分类并对 整个类别进行保存或删除操作, 节省了用户的操作时间, 提升了用户操作 的便利性, 删除不符合质量要求的图像类别可以节省大量的存储空间。
在上述技术方案中, 优选地, 所述分析单元包括: 计算单元, 根据预 设清晰度计算函数计算所述任一图像的清晰度; 以及所述判定单元用于: 在所述任一图像的清晰度小于预设的图像清晰度时, 判定所述任一图像属 于所述第二质量类别, 在所述任一图像的清晰度大于或等于所述预设的图 像清晰度时, 判定所述任一图像属于所述第一质量类别。
在该技术方案中, 可以将图像的清晰度作为图像的质量标准。 通常情 况下, 清晰的图像比不清晰的图像包含更多的信息, 使用评价函数作为预 设清晰度计算函数可以反映表征图像清晰度的一个数值, 在评价函数处于 最大值时的参考图像即清晰度最佳的图像。 目前, 评价清晰度的方法函数 有梯度函数、 频谱函数以及熵函数等。
例如, 釆用梯度函数中的 Tenengrad 函数来计算图像水平和垂直方向 的边缘梯度值, 并用其平方和来设定图像的清晰度范围。 Tenengrad 函数 使用 soble算子来提取边缘的梯度值, 梯度值的平方和为:
Figure imgf000007_0001
y 其中 S ( x,y )是在点 (x,y ) 上与 Soble算子的卷积, 其计算公式为:
Figure imgf000008_0001
其中, G x, 和 (x'_y)分别为图像在横向与纵向上的梯度近似值。 另外, 还可以釆用频谱函数, 使用二维的傅里叶变换对图像的空间频 率进行分析, 以此将图像的高频分量与低频分量分开, 清晰的图像会有更 尖锐、 跳变更强烈的边缘, 包含更多的高频分量, 所以可以通过评估图像 的高频分量来决定图像的清晰度。 但是釆用频谱函数评估清晰度的计算量 很大, 可能影响图像处理的效率。
在上述技术方案中, 优选地, 还包括: 判断单元, 获取所有图像, 按 照所述所有图像的顺序依次判断当前图像与所述当前图像的前一张图像的 相似度; 分组单元, 在所述相似度大于或等于预设相似度时, 将所述当前 图像与所述前一张图像定义为同组图像, 在所述相似度小于所述预设相似 度时, 将所述当前图像与所述前一张图像定义为不同组图像。
在该技术方案中, 由于用户经常会对某一场景拍摄多张质量不同的图 像, 有必要首先判别多个图像是否是属于同一个场景, 然后再对同场景的 图像进行鉴别。 同场景的图像由于其相似度较高, 即使有质量差别, 其清 晰度差别也不会太大, 所以经过初次判断后, 这些图像一般都被归入同一 组别, 如果直接对这些拍摄内容一致的图像进行保存, 会占用大量存储空 间, 而如果直接对这些拍摄内容一致的图像进行删除, 又难保用户不会用 到这些图像。 因此, 还应当对这些内容重复的图像进行再次筛选。
在上述技术方案中, 优选地, 所述判断单元包括: 划分单元, 分别将 所述当前图像和所述前一张图像划分为多个宏块, 获取每个图像的宏块的 RGB值并转换成 YUV亮度和色度分量; 相似度计算单元, 分别计算所述 当前图像和所述前一张图像中对应的宏块的 YUV 分量的均方差值, 并将 所有宏块的 YUV 均方差值进行加和, 以得到所述当前图像和所述前一张 图像的相似度。
在该技术方案中, 可以将图像划分为若干 32x32的宏块, 获取图片宏 块的 RGB值并转换成 YUV亮度和色度分量, 分别计算两幅图片中对应宏 块的 YUV 分量的均方差值, 并将所有宏块的 YUV 均方差值进行加和, 当加和值小于某一阔值时则认为两幅图片比较相似。 在上述技术方案中, 优选地, 还包括: 统计单元, 统计每个所述同组 图像中图像的数量; 以及所述判定单元还用于: 当任一同组图像中图像的 数量为一张时, 判定所述任一同组图像中的图像属于所述第一质量类别, 或者提示用户选择所述任一同组图像中的图像所属的质量类别, 并根据所 述用户的选择判定所述任一同组图像中的图像属于所述第一质量类别或所 述第二质量类别; 所述图像分类装置还包括: 质量比较单元, 当所述任一 同组图像中图像的数量为多张时, 将多张图像的质量进行比较, 并根据比 较结果对所述多张图像进行排名; 所述判定单元还用于: 将所述多张图像 中排名靠前的预设数量的图像判定为属于所述第一质量类别, 将所述多张 图像中的其他图像判定为属于所述第二质量类别。
在该技术方案中, 当对某场景拍摄的图像只有一张时, 即使该图像质 量较低, 用户也可能希望将其存储起来, 因此, 当检测到任一同组图像中 图像的数量为一张时, 可以直接将该组归入第一质量类别, 对该组建议保 存, 也可以向用户提示该组图像的质量较低, 应被归入第二质量类别, 并 为用户提供归入第一质量类别和归入第二质量类别两个选项, 由用户根据 自身需要来决定该组图像的分类。 用户对某一场景拍摄多张内容相同但质 量不同的图像由于其相似度一般被归在同一组, 如果将这样的图像分组直 接保存, 不仅会占用大量空间, 而且保存大量重复内容也没有实际意义, 而如果将这样的图像分组直接删除, 又难保用户不会用到这些图像, 因 此, 可以根据图像质量对该图像分组内的图像进行排序, 设定一个预设数 量, 将排名靠前也就是质量相对较高的预设数量的图像归入第一质量类 别, 将该图像分组内其余排名靠后也就是质量相对较低的图像归入第二质 量类别。 这样, 既可以保证不会丟失对用户可能有用的图像, 也避免了存 储过多内容重复的图像而浪费存储空间。
在上述技术方案中, 优选地, 所述质量比较单元包括: 主角判断单 元, 当所述同一组图像中图像的数量为多张时, 判断所述多张图像的主角 是否为人; 第一检测单元, 在判断结果为所述多张图像的主角为人时, 对 所述多张图像中每张图像的人体图像信息进行检测, 以得到第一检测结 果, 其中, 所述人体图像信息包括人体占据的图像的整体面积和位置信 息、 人脸信息和五官信息; 第一排名单元, 根据所述第一检测结果, 对所 述多张图像进行排名。
在该技术方案中, 当主角为人时, 首先进行人体检测, 根据检测到人 体占据的图像的整体面积和位置判断是否是图像中的主角, 然后再对主角 进行人脸检测和五官检测, 若检测不到人脸和五官, 表明该照片不是用户 想要的照片, 很可能是误拍的照片, 可以把它放到第二质量类别的低质量 照片中备选, 而对于人脸和五官都可以正常检测的照片可以再运用曝光检 测、 锐度检测和噪点检测等维度检测来判断照片的质量, 从而对分组内的 图像进行排名。
在上述技术方案中, 优选地, 所述质量比较单元还包括: 第二检测单 元, 当判断结果为所述多张图像的主角为物时, 对所述多张图像中每张图 像的亮度信息、 锐度信息以及噪点信息进行检测, 以得到第二检测结果; 第二排名单元, 根据所述第二检测结果, 对所述多张图像进行排名。
在该技术方案中, 当多张图像的主角为物时, 对多张图像进行曝光检 测、 锐度检测和噪点检测等维度检测来判断照片的质量, 从而对分组内的 图像进行排名。
在上述技术方案中, 优选地, 还包括: 显示单元, 根据接收到的显示 命令, 对所述第一质量类别中的图像和所述第二质量类别中的图像进行区 分显示。
在该技术方案中, 将第一质量类别中的图像和第二质量类别中的图像 区分显示, 便于用户对两个类别的图像分别进行操作。
在上述技术方案中, 优选地, 还包括: 类别更改单元, 根据接收到的 图像类别更改命令, 将所述任一图像所属的质量类别由所述第一质量类别 更改为所述第二质量类别, 或者由所述第二质量类别更改为所述第一质量 类别。
在该技术方案中, ***对图像判别和分类是完全依据逻辑和算法的, 在***完成对图像的分类后, 用户仍可以根据自身的需求改变图像的分 类, 比如, 用户临时决定保留第二质量类别中的某一图像, 不再对它进行 删除操作, 此时便可以将该图像从第二质量类别调回第一质量类别, 将该 图像保存。
在上述技术方案中, 优选地, 还包括: 删除单元, 根据接收到的删除 命令, 删除所述第二质量类别中的所有图像。
在该技术方案中, 对第二质量类别中的图像进行统一删除, 大大提高 了删除图像的效率, 使用户不用再逐个判断图像质量再删除图像。
通过以上技术方案, 可以根据图像的质量将图像分类处理, 将用户需 要的和不需要的图像区分开来, 对用户不需要的图像统一删除, 使用户操 作简捷方便, 同时还可以节省存储空间。 附图说明
图 1示出了现有技术中的手机相册界面。
图 2示出了根据本发明的实施例的图像分类方法的流程图;
图 3示出了根据本发明的实施例的图像分类装置的框图;
图 4 示出了根据本发明的实施例的清晰图像与模糊图像的对比示意 图;
图 5A和图 5B示出了根据本发明的实施例的手机相册界面。 具体实施方式
为了能够更清楚地理解本发明的上述目的、 特征和优点, 下面结合附 图和具体实施方式对本发明进行进一步的详细描述。 需要说明的是, 在不 冲突的情况下, 本申请的实施例及实施例中的特征可以相互组合。
在下面的描述中阐述了很多具体细节以便于充分理解本发明, 但是, 本发明还可以釆用其他不同于在此描述的其他方式来实施, 因此, 本发明 的保护范围并不受下面公开的具体实施例的限制。
图 2示出了根据本发明的实施例的图像分类方法的流程图。
如图 2所示, 根据本发明的实施例的图像分类方法, 包括:
步骤 202, 对任一图像的质量进行分析;
步骤 204, 根据分析结果判定任一图像所属的质量类别, 其中, 质量 类别包括建议保存的第一质量类别和建议删除的第二质量类别。 在该技术方案中, 可以根据图像的特性 (比如清晰度) 来判断图像的 质量, 通过图像的质量对图像进行分类, 并建议保存符合质量要求的图像 类别, 建议删除不符合质量要求的图像类别。 这样, 就可以对图像进行分 类, 并对整个类别进行保存或删除操作, 节省了用户的操作时间, 提升了 用户操作的便利性, 删除不符合质量要求的图像类别可以节省大量的存储 空间。
在上述技术方案中, 优选地, 步骤 202具体包括: 根据预设清晰度计 算函数计算任一图像的清晰度; 步骤 204具体包括: 在任一图像的清晰度 小于预设的图像清晰度时, 判定任一图像属于第二质量类别; 在任一图像 的清晰度大于或等于预设的图像清晰度时, 判定任一图像属于第一质量类 别。
在该技术方案中, 可以将图像的清晰度作为图像的质量标准。 通常情 况下, 清晰的图像比不清晰的图像包含更多的信息, 使用评价函数作为预 设清晰度计算函数可以反映表征图像清晰度的一个数值, 在评价函数处于 最大值时的参考图像即为清晰度最佳的图像。 目前, 评价清晰度的方法函 数有梯度函数、 频谱函数以及熵函数等。
例如, 釆用梯度函数中的 Tenengrad 函数来计算图像水平和垂直方向 的边缘梯度值, 并用其平方和来设定图像的清晰度范围。 Tenengrad 函数 使用 soble算子来提取边缘的梯度值, 梯度值的平方和为:
Figure imgf000012_0001
y 其中 S ( x,y )是在点 (x,y ) 上与 Soble算子的卷积, 其计算公式为:
Figure imgf000012_0002
其中, Gx (x, 和 (x'_y)分别为图像在横向与纵向上的梯度近似值。 另外, 还可以釆用频谱函数, 使用二维的傅里叶变换对图像的空间频 率进行分析, 以此将图像的高频分量与低频分量分开, 清晰的图像会有更 尖锐、 跳变更强烈的边缘, 包含更多的高频分量, 所以可以通过评估图像 的高频分量来决定图像的清晰度。 但是釆用频谱函数评估清晰度的计算量 很大, 可能影响图像处理的效率。 在上述技术方案中, 优选地, 还包括: 获取所有图像, 按照所有图像 的顺序依次判断当前图像与当前图像的前一张图像的相似度, 在相似度大 于或等于预设相似度时, 将当前图像与前一张图像定义为同组图像, 在相 似度小于预设相似度时, 将当前图像与前一张图像定义为不同组图像。
在该技术方案中, 由于用户经常会对某一场景拍摄多张质量不同的图 像, 有必要首先判别多个图像是否是属于同一个场景, 然后再对同场景的 图像进行鉴别。 同场景的图像由于其相似度较高, 即使有质量差别, 其清 晰度差别也不会太大, 所以经过初次判断后, 这些图像一般都被归入同一 组别, 如果直接对这些拍摄内容一致的图像进行保存, 会占用大量存储空 间, 而如果直接对这些拍摄内容一致的图像进行删除, 又难保用户不会用 到这些图像。 因此, 还应当对这些内容重复的图像进行再次筛选。
在上述技术方案中, 优选地, 判断当前图像与当前图像的前一张图像 的相似度, 具体包括: 分别将当前图像和前一张图像划分为多个宏块, 获 取每个图像的宏块的 RGB值并转换成 YUV亮度和色度分量; 分别计算当 前图像和前一张图像中对应的宏块的 YUV 分量的均方差值, 并将所有宏 块的 YUV均方差值进行加和, 以得到当前图像和前一张图像的相似度。
在该技术方案中, 可以将图像划分为若干 32x32的宏块, 获取图片宏 块的 RGB值并转换成 YUV亮度和色度分量, 分别计算两幅图片中对应宏 块的 YUV 分量的均方差值, 并将所有宏块的 YUV 均方差值进行加和, 当加和值小于某一阔值时则认为两幅图片比较相似。
在上述技术方案中, 优选地, 还包括: 统计每个同组图像中图像的数 量; 以及步骤 204还包括: 当任一同组图像中图像的数量为一张时, 判定 任一同组图像中的图像属于第一质量类别, 或者提示用户选择任一同组图 像中的图像所属的质量类别, 并根据用户的选择判定任一同组图像中的图 像属于第一质量类别或第二质量类别; 当任一同组图像中图像的数量为多 张时, 将多张图像的质量进行比较, 并根据比较结果对多张图像进行排 名; 将多张图像中排名靠前的预设数量的图像判定为属于第一质量类别, 将多张图像中的其他图像判定为属于第二质量类别。
在该技术方案中, 当对某场景拍摄的图像只有一张时, 即使该图像质 量较低, 用户也可能希望将其存储起来, 因此, 当检测到任一同组图像中 图像的数量为一张时, 可以直接将该组归入第一质量类别, 对该组建议保 存, 也可以向用户提示该组图像的质量较低, 应被归入第二质量类别, 并 为用户提供归入第一质量类别和归入第二质量类别两个选项, 由用户根据 自身需要来决定该组图像的分类。 用户对某一场景拍摄多张内容相同但质 量不同的图像由于其相似度一般被归在同一组, 如果将这样的图像分组直 接保存, 不仅会占用大量空间, 而且保存大量重复内容也没有实际意义, 而如果将这样的图像分组直接删除, 又难保用户不会用到这些图像, 因 此, 可以根据图像质量对该图像分组内的图像进行排序, 设定一个预设数 量, 将排名靠前也就是质量相对较高的预设数量的图像归入第一质量类 别, 将该图像分组内其余排名靠后也就是质量相对较低的图像归入第二质 量类别。 这样, 既可以保证不会丟失对用户可能有用的图像, 也避免了存 储过多内容重复的图像而浪费存储空间。
在上述技术方案中, 优选地, 当任一同组图像中图像的数量为多张 时, 将多张图像的质量进行比较, 并根据比较结果对多张图像进行排名, 具体包括: 当同一组图像中图像的数量为多张时, 判断多张图像的主角是 否为人; 在判断结果为多张图像的主角为人时, 对多张图像中每张图像的 人体图像信息进行检测, 以得到第一检测结果, 其中, 人体图像信息包括 人体占据的图像的整体面积和位置信息、 人脸信息和五官信息; 根据第一 检测结果, 对多张图像进行排名。
在该技术方案中, 当主角为人时, 首先进行人体检测, 根据检测到人 体占据的图像的整体面积和位置判断是否是图像中的主角, 然后再对主角 进行人脸检测和五官检测, 若检测不到人脸和五官, 表明该照片不是用户 想要的照片, 很可能是误拍的照片, 可以把它放到第二质量类别的低质量 照片中备选, 而对于人脸和五官都可以正常检测的照片可以再运用曝光检 测、 锐度检测和噪点检测等维度检测来判断照片的质量, 从而对分组内的 图像进行排名。
其中, 曝光检测为通过判断图像的亮度平均值是否大于阔值来判断图 像是否过曝, 通过曝光检测可以将过曝的照片放入第二质量类别建议删 除。 亮度平均值等于照片亮度的总值 /像素点总数与场景曝光系数的乘 积。 场景曝光系数与具体场景有关, 考虑到不同场景本身的亮度值不同, 设计了场景曝光系数的概念, 比如场景为白雪和阳光的情况下本身亮度就 很高, 而场景本身亮度越高, 场景曝光系数就越小, 以避免在白雪和阳光 等场景下得到的亮度平均值超过阔值。 锐度检测是通过对边缘锐度的检测 算法, 统计图像的某一边缘方向的灰度变化情况来评价图像质量, 即灰度 变化越剧烈, 边缘就越清晰, 图像质量就越高。 噪点检测则是指根据含噪 图像与对应均值图像在各像素点的灰色关联系数来识别噪点。
上述技术方案中, 优选地, 当任一同组图像中图像的数量为多张时, 将多张图像的质量进行比较, 并根据比较结果对多张图像进行排名, 还包 括: 当判断结果为多张图像的主角为物时, 对多张图像中每张图像的亮度 信息、 锐度信息以及噪点信息进行检测, 以得到第二检测结果; 根据第二 检测结果, 对多张图像进行排名。
在该技术方案中, 当多张图像的主角为物时, 对多张图像进行曝光检 测、 锐度检测和噪点检测等维度检测来判断照片的质量, 从而对分组内的 图像进行排名。
在上述技术方案中, 优选地, 还包括: 根据接收到的显示命令, 对第 一质量类别中的图像和第二质量类别中的图像进行区分显示。
在该技术方案中, 将第一质量类别中的图像和第二质量类别中的图像 区分显示, 便于用户对两个类别的图像分别进行操作。
在上述技术方案中, 优选地, 还包括: 根据接收到的图像类别更改命 令, 将任一图像所属的质量类别由第一质量类别更改为第二质量类别, 或 者由第二质量类别更改为第一质量类别。
在该技术方案中, ***对图像判别和分类是完全依据逻辑和算法的, 在***完成对图像的分类后, 用户仍可以根据自身的需求改变图像的分 类, 比如, 用户临时决定保留第二质量类别中的某一图像, 不再对它进行 删除操作, 此时便可以将该图像从第二质量类别调回第一质量类别, 将该 图像保存。
在上述技术方案中, 优选地, 还包括: 根据接收到的删除命令, 删除 第二质量类别中的所有图像。
在该技术方案中, 对第二质量类别中的图像进行统一删除, 大大提高 了删除图像的效率, 使用户不用再逐个判断图像质量再删除图像。
图 3示出了根据本发明的实施例的图像分类装置的框图。
如图 3 所示, 根据本发明的实施例的图像分类装置 300, 包括: 分析 单元 302, 对任一图像的质量进行分析; 判定单元 304, 根据分析结果判 定任一图像所属的质量类别, 其中, 质量类别包括建议保存的第一质量类 别和建议删除的第二质量类别。
在该技术方案中, 可以根据图像的特性 (比如清晰度) 来判断图像的 质量, 通过图像的质量对图像进行分类, 并建议保存符合质量要求的图像 类别, 建议删除不符合质量要求的图像类别。 这样, 就可以对图像进行分 类, 并对整个类别进行保存或删除操作, 节省了用户的操作时间, 提升了 用户操作的便利性, 删除不符合质量要求的图像类别可以节省大量的存储 空间。
在上述技术方案中, 优选地, 分析单元 302 包括: 计算单元 3022, 根据预设清晰度计算函数计算任一图像的清晰度; 以及判定单元 304 用 于: 在任一图像的清晰度小于预设的图像清晰度时, 判定任一图像属于第 二质量类别, 在任一图像的清晰度大于或等于预设的图像清晰度时, 判定 任一图像属于第一质量类别。
在该技术方案中, 可以将图像的清晰度作为图像的质量标准。 通常情 况下, 清晰的图像比不清晰的图像包含更多的信息, 使用评价函数作为预 设清晰度计算函数可以反映表征图像清晰度的一个数值, 在评价函数处于 最大值时的参考图像即为清晰度最佳的图像。 目前, 评价清晰度的方法函 数有梯度函数、 频谱函数以及熵函数等。
例如, 釆用梯度函数中的 Tenengrad 函数来计算图像水平和垂直方向 的边缘梯度值, 并用其平方和来设定图像的清晰度范围。 Tenengrad 函数 使用 soble算子来提取边缘的梯度值, 梯度值的平方和为:
Figure imgf000016_0001
y 其中 S ( x,y )是在点 (x,y ) 上与 Soble算子的卷积, 其计算公式为:
Figure imgf000017_0001
其中, G x, 和 (x'_y)分别为图像在横向与纵向上的梯度近似值。 另外, 还可以釆用频谱函数, 使用二维的傅里叶变换对图像的空间频 率进行分析, 以此将图像的高频分量与低频分量分开, 清晰的图像会有更 尖锐、 跳变更强烈的边缘, 包含更多的高频分量, 所以可以通过评估图像 的高频分量来决定图像的清晰度。 但是釆用频谱函数评估清晰度的计算量 很大, 可能影响图像处理的效率。
在上述技术方案中, 优选地, 还包括: 判断单元 306, 获取所有图 像, 按照所有图像的顺序依次判断当前图像与当前图像的前一张图像的相 似度; 分组单元 308, 在相似度大于或等于预设相似度时, 将当前图像与 前一张图像定义为同组图像, 在相似度小于预设相似度时, 将当前图像与 前一张图像定义为不同组图像。
在该技术方案中, 由于用户经常会对某一场景拍摄多张质量不同的图 像, 有必要首先判别多个图像是否是属于同一个场景, 然后再对同场景的 图像进行鉴别。 同场景的图像由于其相似度较高, 即使有质量差别, 其清 晰度差别也不会太大, 所以经过初次判断后, 这些图像一般都被归入同一 组别, 如果直接对这些拍摄内容一致的图像进行保存, 会占用大量存储空 间, 而如果直接对这些拍摄内容一致的图像进行删除, 又难保用户不会用 到这些图像。 因此, 还应当对这些内容重复的图像进行再次筛选。
在上述技术方案中, 优选地, 判断单元 306 包括: 划分单元 3062, 分别将当前图像和前一张图像划分为多个宏块, 获取每个图像的宏块的 RGB值并转换成 YUV亮度和色度分量; 相似度计算单元 3064, 分别计算 当前图像和前一张图像中对应的宏块的 YUV 分量的均方差值, 并将所有 宏块的 YUV 均方差值进行加和, 以得到当前图像和前一张图像的相似 度。
在该技术方案中, 可以将图像划分为若干 32x32的宏块, 获取图片宏 块的 RGB值并转换成 YUV亮度和色度分量, 分别计算两幅图片中对应宏 块的 YUV 分量的均方差值, 并将所有宏块的 YUV 均方差值进行加和, 当加和值小于某一阔值时则认为两幅图片比较相似。 在上述技术方案中, 优选地, 还包括: 统计单元 310, 统计每个同组 图像中图像的数量; 以及判定单元 304还用于: 当任一同组图像中图像的 数量为一张时, 判定任一同组图像中的图像属于第一质量类别, 或者提示 用户选择任一同组图像中的图像所属的质量类别, 并根据用户的选择判定 任一同组图像中的图像属于第一质量类别或第二质量类别; 图像分类装置 300 还包括: 质量比较单元 312, 当任一同组图像中图像的数量为多张 时, 将多张图像的质量进行比较, 并根据比较结果对多张图像进行排名; 判定单元 304还用于: 将多张图像中排名靠前的预设数量的图像判定为属 于第一质量类别, 将多张图像中的其他图像判定为属于第二质量类别。
在该技术方案中, 当对某场景拍摄的图像只有一张时, 即使该图像质 量较低, 用户也可能希望将其存储起来, 因此, 当检测到任一同组图像中 图像的数量为一张时, 可以直接将该组归入第一质量类别, 对该组建议保 存, 也可以向用户提示该组图像的质量较低, 应被归入第二质量类别, 并 为用户提供归入第一质量类别和归入第二质量类别两个选项, 由用户根据 自身需要来决定该组图像的分类。 用户对某一场景拍摄多张内容相同但质 量不同的图像由于其相似度一般被归在同一组, 如果将这样的图像分组直 接保存, 不仅会占用大量空间, 而且保存大量重复内容也没有实际意义, 而如果将这样的图像分组直接删除, 又难保用户不会用到这些图像, 因 此, 可以根据图像质量对该图像分组内的图像进行排序, 设定一个预设数 量, 将排名靠前也就是质量相对较高的预设数量的图像归入第一质量类 别, 将该图像分组内其余排名靠后也就是质量相对较低的图像归入第二质 量类别。 这样, 既可以保证不会丟失对用户可能有用的图像, 也避免了存 储过多内容重复的图像而浪费存储空间。
在上述技术方案中, 优选地, 质量比较单元 312 包括: 主角判断单元 3122 , 当同一组图像中图像的数量为多张时, 判断多张图像的主角是否为 人; 第一检测单元 3124, 在判断结果为多张图像的主角为人时, 对多张 图像中每张图像的人体图像信息进行检测, 以得到第一检测结果, 其中, 人体图像信息包括人体占据的图像的整体面积和位置信息、 人脸信息和五 官信息; 第一排名单元, 根据第一检测结果, 对多张图像进行排名。 在该技术方案中, 当主角为人时, 首先进行人体检测, 根据检测到人 体占据的图像的整体面积和位置判断是否是图像中的主角, 然后再对主角 进行人脸检测和五官检测, 若检测不到人脸和五官, 表明该照片不是用户 想要的照片, 很可能是误拍的照片, 可以把它放到第二质量类别的低质量 照片中备选, 而对于人脸和五官都可以正常检测的照片可以再运用曝光检 测、 锐度检测和噪点检测等维度检测来判断照片的质量, 从而对分组内的 图像进行排名。
在上述技术方案中, 优选地, 质量比较单元 312还包括: 第二检测单 元 3126, 当判断结果为多张图像的主角为物时, 对多张图像中每张图像 的亮度信息、 锐度信息以及噪点信息进行检测, 以得到第二检测结果; 第 二排名单元, 根据第二检测结果, 对多张图像进行排名。
在该技术方案中, 当多张图像的主角为物时, 对多张图像进行曝光检 测、 锐度检测和噪点检测等维度检测来判断照片的质量, 从而对分组内的 图像进行排名。
在上述技术方案中, 优选地, 还包括: 显示单元 314, 根据接收到的 显示命令, 对第一质量类别中的图像和第二质量类别中的图像进行区分显 示。
在该技术方案中, 将第一质量类别中的图像和第二质量类别中的图像 区分显示, 便于用户对两个类别的图像分别进行操作。
在上述技术方案中, 优选地, 还包括: 类别更改单元 316, 根据接收 到的图像类别更改命令, 将任一图像所属的质量类别由第一质量类别更改 为第二质量类别, 或者由第二质量类别更改为第一质量类别。
在该技术方案中, ***对图像判别和分类是完全依据逻辑和算法的, 在***完成对图像的分类后, 用户仍可以根据自身的需求改变图像的分 类, 比如, 用户临时决定保留第二质量类别中的某一图像, 不再对它进行 删除操作, 此时便可以将该图像从第二质量类别调回第一质量类别, 将该 图像保存。
在上述技术方案中, 优选地, 还包括: 删除单元 318, 根据接收到的 删除命令, 删除第二质量类别中的所有图像。 在该技术方案中, 对第二质量类别中的图像进行统一删除, 大大提高 了删除图像的效率, 使用户不用再逐个判断图像质量再删除图像。
图 4 示出了根据本发明的实施例的清晰图像与模糊图像的对比示意 图。
如图 4所示, 在经过放大图像后可以发现, 清晰图像和模糊图像的清 晰度差异很大, 清晰度较高的图像质量更好, 因此, 可以将图像的清晰度 作为图像的质量标准, 对图像进行分类。
根据局部放大后的清晰图像和模糊图像可知, 通常情况下, 清晰的图 像比不清晰的图像包含更多的信息, 使用评价函数作为预设清晰度计算函 数可以反映表征图像清晰度的一个数值, 在评价函数处于最大值时的参考 图像即清晰度最佳的图像。 目前, 评价清晰度的方法函数有梯度函数、 频 谱函数以及熵函数。
例如, 釆用梯度函数中的 Tenengrad 函数来计算图像水平和垂直方向 的边缘梯度值, 并用其平方和来设定图像的清晰度范围。 Tenengrad 函数 使用 soble算子来提取边缘的梯度值, 梯度值的平方和为:
Figure imgf000020_0001
y 其中 S ( x,y )是在点 (x,y ) 上与 Soble算子的卷积, 其计算公式为:
Figure imgf000020_0002
其中, Gx (x, 和 (x'_y)分别为图像在横向与纵向上的梯度近似值。 另外, 还可以釆用频谱函数, 使用二维的傅里叶变换对图像的空间频 率进行分析, 以此将图像的高频分量与低频分量分开, 清晰的图像会有更 尖锐、 跳变更强烈的边缘, 包含更多的高频分量, 所以可以通过评估图像 的高频分量来决定图像的清晰度。 但是釆用频谱函数评估清晰度的计算量 很大, 可能影响图像处理的效率。
根据清晰度进行了图像质量评价以后, 就可以将判别出的低质量图像 提取出来, 归为 "低质量照片" 一类。 这样, 批量地删除不符合质量要求 的 "低质量照片" 既可以使用户操作简便, 还可以节省大量的存储空间。
图 5A和图 5B示出了根据本 '明的实施例的手机相册界面。 如图 5A 所示, 在手机相册界面设置 "所有图像" 选项和 "低质量图 像" 选项。 低质量图像的搜索过程和鉴别过程可以通过两种方式来实现, 可以对拍摄的所有图像进行批量处理, 在相册中对每一张图像分别进行鉴 别, 分出低质量图像组, 也可以每拍一张图像就鉴别一次, 并将低质量图 像进行归类归入低质量图像组。 这样一种搜索和鉴别方法可以在相册的后 台处理完成。 处理完毕后, 可以在手机相册界面中显示 "所有图像" 选项 和 "低质量图像" 选项, 这样就可以给用户一个选择, 让用户自己选择是 查看所有图像还是只查看低质量图像。
当用户选择 "低质量图像" 选项后, 就可以得到如图 5B 示出的低质 量图像界面, 该界面只显示手机自动鉴别出的低质量图像, 让其展示在用 户眼前, 让用户自行决定是否将其中的某些图像删除。
以上结合附图详细说明了本发明的技术方案, 通过本发明的技术方 案, 可以根据图像的质量将图像分类处理, 将用户需要的和不需要的图像 区分开来, 对用户不需要的图像统一删除, 使用户操作简捷方便, 同时还 可以节省存储空间。
根据本发明的实施方式, 还提供了一种存储在非易失性机器可读介质 上的程序产品, 用于终端中的图像分类, 所述程序产品包括用于使计算机 ***执行以下步骤的机器可执行指令: 对任一图像的质量进行分析; 根据 分析结果判定所述任一图像所属的质量类别, 其中, 所述质量类别包括建 议保存的第一质量类别和建议删除的第二质量类别。
根据本发明的实施方式, 还提供了一种非易失机器可读介质, 存储有 用于终端中图像分类的程序产品, 所述程序产品包括用于使计算机***执 行以下步骤的机器可执行指令: 对任一图像的质量进行分析; 根据分析结 果判定所述任一图像所属的质量类别, 其中, 所述质量类别包括建议保存 的第一质量类别和建议删除的第二质量类别。
根据本发明的实施方式, 还提供了一种机器可读程序, 所述程序使机 器执行如上所述技术方案中任一所述的图像分类方法。
根据本发明的实施方式, 还提供了一种存储有机器可读程序的存储介 质, 其中, 所述机器可读程序使得机器执行如上所述技术方案中任一所述 的图像分类方法。 在本发明中, 术语 "第一" 、 "第二" 仅用于描述的目的, 而不能理 解为指示或暗示相对重要性; 术语 "多个" 表示两个或两个以上。 对于本 领域的普通技术人员而言, 可以根据具体情况理解上述术语在本发明中的 具体含义。
以上所述仅为本发明的优选实施例而已, 并不用于限制本发明, 对于 本领域的技术人员来说, 本发明可以有各种更改和变化。 凡在本发明的精 神和原则之内, 所作的任何修改、 等同替换、 改进等, 均应包含在本发明 的保护范围之内。

Claims

权 利 要 求 书
1. 一种图像分类方法, 其特征在于, 包括:
分析步骤, 对任一图像的质量进行分析;
判定步骤, 根据分析结果判定所述任一图像所属的质量类别, 其中, 所述质量类别包括建议保存的第一质量类别和建议删除的第二质量类别。
2. 根据权利要求 1所述的图像分类方法, 其特征在于,
所述分析步骤具体包括:
根据预设清晰度计算函数计算所述任一图像的清晰度;
所述判定步骤具体包括:
在所述任一图像的清晰度小于预设的图像清晰度时, 判定所述任一图 像属于所述第二质量类别;
在所述任一图像的清晰度大于或等于所述预设的图像清晰度时, 判定 所述任一图像属于所述第一质量类别。
3. 根据权利要求 2所述的图像分类方法, 其特征在于, 还包括: 获取所有图像, 按照所述所有图像的顺序依次判断当前图像与所述当 前图像的前一张图像的相似度, 在所述相似度大于或等于预设相似度时, 将所述当前图像与所述前一张图像定义为同组图像, 在所述相似度小于所 述预设相似度时, 将所述当前图像与所述前一张图像定义为不同组图像。
4. 根据权利要求 3 所述的图像分类方法, 其特征在于, 判断当前图 像与所述当前图像的前一张图像的相似度, 具体包括:
分别将所述当前图像和所述前一张图像划分为多个宏块, 获取每个图 像的宏块的 RGB值并转换成 YUV亮度和色度分量;
分别计算所述当前图像和所述前一张图像中对应的宏块的 YUV 分量 的均方差值, 并将所有宏块的 YUV 均方差值进行加和, 以得到所述当前 图像和所述前一张图像的相似度。
5. 根据权利要求 3所述的图像分类方法, 其特征在于, 还包括: 统计每个所述同组图像中图像的数量;
当任一同组图像中图像的数量为一张时, 判定所述任一同组图像中的 图像属于所述第一质量类别, 或者
提示用户选择所述任一同组图像中的图像所属的质量类别, 并根据所 述用户的选择判定所述任一同组图像中的图像属于所述第一质量类别或所 述第二质量类别;
当所述任一同组图像中图像的数量为多张时, 将多张图像的质量进行 比较, 并根据比较结果对所述多张图像进行排名;
将所述多张图像中排名靠前的预设数量的图像判定为属于所述第一质 量类别, 将所述多张图像中的其他图像判定为属于所述第二质量类别。
6. 根据权利要求 5 所述的图像分类方法, 其特征在于, 当所述任一 同组图像中图像的数量为多张时, 将多张图像的质量进行比较, 并根据比 较结果对所述多张图像进行排名, 具体包括:
当所述同一组图像中图像的数量为多张时, 判断所述多张图像的主角 是否为人;
在判断结果为所述多张图像的主角为人时, 对所述多张图像中每张图 像的人体图像信息进行检测, 以得到第一检测结果, 其中, 所述人体图像 信息包括人体占据的图像的整体面积和位置信息、 人脸信息和五官信息; 根据所述第一检测结果, 对所述多张图像进行排名。
7. 根据权利要求 6 所述的图像分类方法, 其特征在于, 当所述任一 同组图像中图像的数量为多张时, 将多张图像的质量进行比较, 并根据比 较结果对所述多张图像进行排名, 还包括:
当判断结果为所述多张图像的主角为物时, 对所述多张图像中每张图 像的亮度信息、 锐度信息以及噪点信息进行检测, 以得到第二检测结果; 根据所述第二检测结果, 对所述多张图像进行排名。
8. 根据权利要求 1所述的图像分类方法, 其特征在于, 还包括: 根据接收到的显示命令, 对所述第一质量类别中的图像和所述第二质 量类别中的图像进行区分显示。
9. 根据权利要求 1所述的图像分类方法, 其特征在于, 还包括: 根据接收到的图像类别更改命令, 将所述任一图像所属的质量类别由 所述第一质量类别更改为所述第二质量类别, 或者由所述第二质量类别更 改为所述第一质量类别。
10. 根据权利要求 1 至 9 中任一项所述的图像分类方法, 其特征在 于, 还包括:
根据接收到的删除命令, 删除所述第二质量类别中的所有图像。
11. 一种图像分类装置, 其特征在于, 包括:
分析单元, 对任一图像的质量进行分析;
判定单元, 根据分析结果判定所述任一图像所属的质量类别, 其中, 所述质量类别包括建议保存的第一质量类别和建议删除的第二质量类别。
12. 根据权利要求 11所述的图像分类装置, 其特征在于,
所述分析单元包括:
计算单元, 根据预设清晰度计算函数计算所述任一图像的清晰度; 以 及
所述判定单元用于:
在所述任一图像的清晰度小于预设的图像清晰度时, 判定所述任一图 像属于所述第二质量类别,
在所述任一图像的清晰度大于或等于所述预设的图像清晰度时, 判定 所述任一图像属于所述第一质量类别。
13. 根据权利要求 12所述的图像分类装置, 其特征在于, 还包括: 判断单元, 获取所有图像, 按照所述所有图像的顺序依次判断当前图 像与所述当前图像的前一张图像的相似度;
分组单元, 在所述相似度大于或等于预设相似度时, 将所述当前图像 与所述前一张图像定义为同组图像, 在所述相似度小于所述预设相似度 时, 将所述当前图像与所述前一张图像定义为不同组图像。
14. 根据权利要求 13 所述的图像分类装置, 其特征在于, 所述判断 单元包括:
划分单元, 分别将所述当前图像和所述前一张图像划分为多个宏块, 获取每个图像的宏块的 RGB值并转换成 YUV亮度和色度分量;
相似度计算单元, 分别计算所述当前图像和所述前一张图像中对应的 宏块的 YUV 分量的均方差值, 并将所有宏块的 YUV 均方差值进行加 和, 以得到所述当前图像和所述前一张图像的相似度。
15. 根据权利要求 13所述的图像分类装置, 其特征在于, 还包括: 统计单元, 统计每个所述同组图像中图像的数量; 以及
所述判定单元还用于:
当任一同组图像中图像的数量为一张时, 判定所述任一同组图像中的 图像属于所述第一质量类别, 或者
提示用户选择所述任一同组图像中的图像所属的质量类别, 并根据所 述用户的选择判定所述任一同组图像中的图像属于所述第一质量类别或所 述第二质量类别;
所述图像分类装置还包括:
质量比较单元, 当所述任一同组图像中图像的数量为多张时, 将多张 图像的质量进行比较, 并根据比较结果对所述多张图像进行排名;
所述判定单元还用于:
将所述多张图像中排名靠前的预设数量的图像判定为属于所述第一质 量类别, 将所述多张图像中的其他图像判定为属于所述第二质量类别。
16. 根据权利要求 15 所述的图像分类装置, 其特征在于, 所述质量 比较单元包括:
主角判断单元, 当所述同一组图像中图像的数量为多张时, 判断所述 多张图像的主角是否为人;
第一检测单元, 在判断结果为所述多张图像的主角为人时, 对所述多 张图像中每张图像的人体图像信息进行检测, 以得到第一检测结果, 其 中, 所述人体图像信息包括人体占据的图像的整体面积和位置信息、 人脸 信息和五官信息;
第一排名单元, 根据所述第一检测结果, 对所述多张图像进行排名。
17. 根据权利要求 16 所述的图像分类装置, 其特征在于, 所述质量 比较单元还包括:
第二检测单元, 当判断结果为所述多张图像的主角为物时, 对所述多 张图像中每张图像的亮度信息、 锐度信息以及噪点信息进行检测, 以得到 第二检测结果; 第二排名单元, 根据所述第二检测结果, 对所述多张图像进行排名。
18. 根据权利要求 11所述的图像分类装置, 其特征在于, 还包括: 显示单元, 根据接收到的显示命令, 对所述第一质量类别中的图像和 所述第二质量类别中的图像进行区分显示。
19. 根据权利要求 11所述的图像分类装置, 其特征在于, 还包括: 类别更改单元, 根据接收到的图像类别更改命令, 将所述任一图像所 属的质量类别由所述第一质量类别更改为所述第二质量类别, 或者由所述 第二质量类别更改为所述第一质量类别。
20. 根据权利要求 11至 19中任一项所述的图像分类装置, 其特征在 于, 还包括:
删除单元, 根据接收到的删除命令, 删除所述第二质量类别中的所有 图像。
PCT/CN2014/082657 2014-07-21 2014-07-21 图像分类方法和图像分类装置 WO2016011589A1 (zh)

Priority Applications (4)

Application Number Priority Date Filing Date Title
CN201480080711.2A CN106575223B (zh) 2014-07-21 2014-07-21 图像分类方法和图像分类装置
EP14898089.9A EP3196758B1 (en) 2014-07-21 2014-07-21 Image classification method and image classification apparatus
PCT/CN2014/082657 WO2016011589A1 (zh) 2014-07-21 2014-07-21 图像分类方法和图像分类装置
US15/327,626 US10289939B2 (en) 2014-07-21 2014-07-21 Image classification method and image classification apparatus

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2014/082657 WO2016011589A1 (zh) 2014-07-21 2014-07-21 图像分类方法和图像分类装置

Publications (1)

Publication Number Publication Date
WO2016011589A1 true WO2016011589A1 (zh) 2016-01-28

Family

ID=55162394

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2014/082657 WO2016011589A1 (zh) 2014-07-21 2014-07-21 图像分类方法和图像分类装置

Country Status (4)

Country Link
US (1) US10289939B2 (zh)
EP (1) EP3196758B1 (zh)
CN (1) CN106575223B (zh)
WO (1) WO2016011589A1 (zh)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108241821A (zh) * 2016-12-23 2018-07-03 北京三星通信技术研究有限公司 图像处理设备和方法
CN108804653A (zh) * 2018-06-07 2018-11-13 联想(北京)有限公司 用于图片处理的方法、装置以及电子设备

Families Citing this family (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10509770B2 (en) 2015-07-13 2019-12-17 Samsung Electronics Co., Ltd. Heuristic interface for enabling a computer device to utilize data property-based data placement inside a nonvolatile memory device
US10282324B2 (en) 2015-07-13 2019-05-07 Samsung Electronics Co., Ltd. Smart I/O stream detection based on multiple attributes
US11461010B2 (en) 2015-07-13 2022-10-04 Samsung Electronics Co., Ltd. Data property-based data placement in a nonvolatile memory device
US10296264B2 (en) 2016-02-09 2019-05-21 Samsung Electronics Co., Ltd. Automatic I/O stream selection for storage devices
US11748877B2 (en) * 2017-05-11 2023-09-05 The Research Foundation For The State University Of New York System and method associated with predicting segmentation quality of objects in analysis of copious image data
US10901907B2 (en) 2017-10-19 2021-01-26 Samsung Electronics Co., Ltd. System and method for identifying hot data and stream in a solid-state drive
CN108875518A (zh) * 2017-12-15 2018-11-23 北京旷视科技有限公司 图像处理与图像分类方法、装置和***及存储介质
CN108109145A (zh) * 2018-01-02 2018-06-01 中兴通讯股份有限公司 图像质量检测方法、装置、存储介质和电子装置
CN114025104B (zh) * 2021-11-03 2024-02-20 江苏金视传奇科技有限公司 一种彩色图像显示装置及彩色图像显示***
CN116843683B (zh) * 2023-08-30 2024-03-05 荣耀终端有限公司 设备成像清晰度评估方法、***和装置

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101170940A (zh) * 2005-05-20 2008-04-30 奥林巴斯医疗株式会社 图像显示装置
CN101414348A (zh) * 2007-10-19 2009-04-22 三星电子株式会社 多角度人脸识别方法和***
US20110080616A1 (en) * 2009-10-07 2011-04-07 Susan Yang Automatic Red-Eye Object Classification In Digital Photographic Images
CN103412753A (zh) * 2013-08-12 2013-11-27 宇龙计算机通信科技(深圳)有限公司 图片删除方法及装置

Family Cites Families (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6535636B1 (en) * 1999-03-23 2003-03-18 Eastman Kodak Company Method for automatically detecting digital images that are undesirable for placing in albums
US6301440B1 (en) 2000-04-13 2001-10-09 International Business Machines Corp. System and method for automatically setting image acquisition controls
US7809197B2 (en) * 2004-12-09 2010-10-05 Eastman Kodak Company Method for automatically determining the acceptability of a digital image
US8086046B2 (en) * 2005-06-27 2011-12-27 Pioneer Corporation Image analysis device and image analysis method
CN101860680B (zh) * 2009-04-08 2011-12-21 鸿富锦精密工业(深圳)有限公司 影像处理装置及其影像处理方法
US8346767B2 (en) * 2010-04-21 2013-01-01 Microsoft Corporation Image search result summarization with informative priors
CN101853504B (zh) * 2010-05-07 2012-04-25 厦门大学 基于视觉特性与结构相似度的图像质量评测方法
US8635172B1 (en) * 2011-10-07 2014-01-21 Google Inc. Dynamic techniques for evaluating quality of clustering or classification system aimed to minimize the number of manual reviews based on Bayesian inference and Markov Chain Monte Carlo (MCMC) techniques
CN103369209B (zh) * 2013-07-31 2016-08-17 上海通途半导体科技有限公司 视频降噪装置及方法

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101170940A (zh) * 2005-05-20 2008-04-30 奥林巴斯医疗株式会社 图像显示装置
CN101414348A (zh) * 2007-10-19 2009-04-22 三星电子株式会社 多角度人脸识别方法和***
US20110080616A1 (en) * 2009-10-07 2011-04-07 Susan Yang Automatic Red-Eye Object Classification In Digital Photographic Images
CN103412753A (zh) * 2013-08-12 2013-11-27 宇龙计算机通信科技(深圳)有限公司 图片删除方法及装置

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
See also references of EP3196758A4 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108241821A (zh) * 2016-12-23 2018-07-03 北京三星通信技术研究有限公司 图像处理设备和方法
CN108804653A (zh) * 2018-06-07 2018-11-13 联想(北京)有限公司 用于图片处理的方法、装置以及电子设备

Also Published As

Publication number Publication date
CN106575223B (zh) 2020-05-19
EP3196758A4 (en) 2018-03-28
EP3196758A1 (en) 2017-07-26
US20170308772A1 (en) 2017-10-26
US10289939B2 (en) 2019-05-14
EP3196758B1 (en) 2021-07-07
CN106575223A (zh) 2017-04-19

Similar Documents

Publication Publication Date Title
WO2016011589A1 (zh) 图像分类方法和图像分类装置
CN105072337B (zh) 图片处理方法及装置
AU2017261537B2 (en) Automated selection of keeper images from a burst photo captured set
US10147163B2 (en) Systems and methods for automated image cropping
WO2016065999A1 (zh) 相册管理方法和装置
CN108174185B (zh) 一种拍照方法、装置及终端
TW201911130A (zh) 一種翻拍影像識別方法及裝置
WO2017000716A2 (zh) 图片的管理方法、装置及终端设备
WO2015058381A1 (zh) 从连拍图像中选择图像的方法和终端
CN109740444B (zh) 人流量信息展示方法及相关产品
WO2016127478A1 (zh) 一种图像处理方法、装置和终端
WO2022111072A1 (zh) 图像拍摄方法、装置、电子设备、服务器及存储介质
WO2021004186A1 (zh) 一种人脸采集方法、装置、***、设备及介质
JP2017076909A (ja) 画像処理装置、画像処理方法、及びプログラム
WO2017143852A1 (zh) 一种图片处理方法、装置及电子设备
US20160070960A1 (en) Image processing apparatus
CN106095876B (zh) 图像处理方法及装置
US10296539B2 (en) Image extraction system, image extraction method, image extraction program, and recording medium storing program
US10657407B2 (en) Image processing apparatus, image processing method, and recording medium
WO2015196681A1 (zh) 一种图片处理方法及电子设备
WO2017177559A1 (zh) 一种图像管理方法和装置
KR101748139B1 (ko) 복수의 카메라가 촬영한 영상을 관리하는 영상 처리 장치 및 방법
CN108174091B (zh) 图像处理方法、装置、存储介质及电子设备
CN116958795A (zh) 翻拍图像的识别方法、装置、电子设备及存储介质
CN106402717B (zh) 一种ar播放控制方法及智能台灯

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: 14898089

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

REEP Request for entry into the european phase

Ref document number: 2014898089

Country of ref document: EP

WWE Wipo information: entry into national phase

Ref document number: 2014898089

Country of ref document: EP

WWE Wipo information: entry into national phase

Ref document number: 15327626

Country of ref document: US