WO2017156864A1 - 图像识别方法、装置、设备及非易失性计算机存储介质 - Google Patents

图像识别方法、装置、设备及非易失性计算机存储介质 Download PDF

Info

Publication number
WO2017156864A1
WO2017156864A1 PCT/CN2016/082969 CN2016082969W WO2017156864A1 WO 2017156864 A1 WO2017156864 A1 WO 2017156864A1 CN 2016082969 W CN2016082969 W CN 2016082969W WO 2017156864 A1 WO2017156864 A1 WO 2017156864A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
difference area
feature
obtaining
template
Prior art date
Application number
PCT/CN2016/082969
Other languages
English (en)
French (fr)
Inventor
刘国翌
李广
Original Assignee
百度在线网络技术(北京)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 百度在线网络技术(北京)有限公司 filed Critical 百度在线网络技术(北京)有限公司
Priority to US16/080,945 priority Critical patent/US11455783B2/en
Publication of WO2017156864A1 publication Critical patent/WO2017156864A1/zh

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00Arrangements for image or video recognition or understanding
    • G06V10/20Image preprocessing
    • G06V10/25Determination of region of interest [ROI] or a volume of interest [VOI]
    • 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
    • G06F18/24147Distances to closest patterns, e.g. nearest neighbour classification
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/50Information retrieval; Database structures therefor; File system structures therefor of still image data
    • G06F16/58Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
    • G06F16/583Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00Pattern recognition
    • G06F18/20Analysing
    • G06F18/21Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • 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
    • G06F18/24133Distances to prototypes
    • G06F18/24143Distances to neighbourhood prototypes, e.g. restricted Coulomb energy networks [RCEN]
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00Pattern recognition
    • G06F18/20Analysing
    • G06F18/28Determining representative reference patterns, e.g. by averaging or distorting; Generating dictionaries
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T3/00Geometric image transformations in the plane of the image
    • G06T3/02Affine transformations
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00Arrangements for image or video recognition or understanding
    • G06V10/70Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/764Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00Arrangements for image or video recognition or understanding
    • G06V10/70Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/77Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
    • G06V10/772Determining representative reference patterns, e.g. averaging or distorting patterns; Generating dictionaries
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00Arrangements for image or video recognition or understanding
    • G06V10/70Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/82Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V2201/00Indexing scheme relating to image or video recognition or understanding
    • G06V2201/09Recognition of logos

Definitions

  • the present invention relates to image processing technologies, and in particular, to an image recognition method, apparatus, device, and nonvolatile computer storage medium.
  • Deep learning requires a high number of training samples, often in the hundreds of thousands or even millions of training samples.
  • aspects of the present invention provide an image recognition method, apparatus, device, and non-volatile computer storage medium for performing recognition processing on a limited number of images.
  • An aspect of the present invention provides an image recognition method, including:
  • any possible implementation manner further provide an implementation manner of obtaining an image to be identified of a specified size, including:
  • the obtained image of any size to be recognized is adjusted to the image to be recognized of the specified size by affine transformation.
  • the difference area image is extracted from the image to be recognized according to a pre-designated area position.
  • the obtaining the image feature of the different area image according to the difference area image including:
  • An image feature of the different area image is obtained from the difference area image using a model trained by the universal data set.
  • the method further includes:
  • Template features for each of the specified categories are obtained based on the template region images for each of the specified categories.
  • an image recognition apparatus comprising:
  • An obtaining unit configured to acquire an image to be identified of a specified size
  • An extracting unit configured to extract a difference area image from the image to be identified
  • a feature unit configured to obtain an image feature of the difference area image according to the difference area image
  • an identifying unit configured to obtain, according to the image feature of the different area image and the preset template feature, a recognition result of the image to be recognized.
  • any possible implementation manner further provide an implementation manner, where the acquiring unit is specifically configured to
  • the obtained image of any size to be recognized is adjusted to the image to be recognized of the specified size by affine transformation.
  • the difference area image is extracted from the image to be recognized according to a pre-designated area position.
  • An image feature of the different area image is obtained from the difference area image using a model trained by the universal data set.
  • the feature unit is further used for
  • Template features for each of the specified categories are obtained based on the template region images for each of the specified categories.
  • an apparatus comprising:
  • One or more processors are One or more processors;
  • One or more programs the one or more programs being stored in the memory, when executed by the one or more processors:
  • a nonvolatile computer storage medium storing one or more programs when the one or more programs are executed by a device causes The device:
  • the embodiment of the present invention obtains an image to be identified of a specified size, and further extracts a difference area image from the image to be identified, and obtains an image feature of the image of the difference area according to the image of the difference area.
  • Obtaining a recognition result of the image to be recognized according to an image feature of the difference area image and a preset template feature which can be implemented without using a deep learning method based on hundreds of thousands or even millions of training samples. Identify and process images with a limited number of categories.
  • the training samples are trained to obtain a model by using a deep learning method. It is a model that can be trained by using a universal data set, so that the workload of data acquisition and model training of large-scale training samples is removed, and the algorithm development time of image recognition processing can be effectively accelerated.
  • the accuracy of the image recognition processing can be effectively ensured by manually specifying the location of the region having a large difference.
  • FIG. 1 is a schematic flowchart of an image recognition method according to an embodiment of the present invention.
  • FIG. 2 is a schematic structural diagram of an image recognition apparatus according to another embodiment of the present invention.
  • FIG. 3 is a schematic diagram of an image of a different area in the embodiment corresponding to FIG. 1.
  • FIG. 3 is a schematic diagram of an image of a different area in the embodiment corresponding to FIG. 1.
  • the terminal involved in the embodiment of the present invention may include but is not limited to a hand.
  • PDA Personal Digital Assistant
  • wireless handheld device tablet computer
  • PC personal computer
  • MP3 player MP4 player
  • wearable device for example, smart glasses, Smart watches, smart bracelets, etc.
  • FIG. 1 is a schematic flowchart of an image recognition method according to an embodiment of the present invention, as shown in FIG. 1 .
  • the so-called image refers to a certain image format
  • the image data that is, the pixels of the image are stored in a certain manner
  • the formed file may also be referred to as an image file.
  • the image format of the image may include, but is not limited to, a bitmap (BMP) format, a Portable Network Graphic Format (PNG), and a Joint Photographic Experts Group (Joint Photographic Experts Group, The JPEG) format and the Exchangeable Image File Format (EXIF) are not particularly limited in this embodiment.
  • BMP bitmap
  • PNG Portable Network Graphic Format
  • JPEG Joint Photographic Experts Group
  • EXIF Exchangeable Image File Format
  • execution entities of 101 to 104 may be applications located at the local terminal. Or it may be a plug-in or a software development kit (SDK) set in an application located in the local terminal, or may be a processing engine located in the network side server, or may be located on the network side.
  • SDK software development kit
  • the distributed system is not particularly limited in this embodiment.
  • the application may be a local application (nativeApp) installed on the terminal, or may be a web application (webApp) of the browser on the terminal, which is not specifically limited in this embodiment.
  • the image feature of the image and the preset template feature obtain the recognition result of the image to be recognized, and the image with limited number of categories can be realized without using the deep learning method based on hundreds of thousands or even millions of training samples. deal with.
  • the image to be identified may be acquired by using an image sensor.
  • the image sensor may be a Charge Coupled Device (CCD) sensor, or may be a Metal Oxide Semiconductor (CMOS) sensor, which is not particularly limited in this embodiment.
  • CCD Charge Coupled Device
  • CMOS Metal Oxide Semiconductor
  • the captured image contains, in addition to the target object corresponding to the image to be recognized, some other objects are usually included as background images, for example, an image of a person holding a renminbi in the hand, in addition to the image to be recognized
  • the corresponding target object may also contain other objects such as human hands and checkout counters as background images, etc. Therefore, it is necessary to further adopt the traditional image detection method, for example, scale-invariant feature transformation. (Scale-Invariant Feature Transform, SIFT) algorithm or the like, finds an area of a target object in an image as the image to be recognized.
  • SIFT Scale-Invariant Feature Transform
  • the obtained image of any size to be recognized may be specifically adjusted to the image to be identified of the specified size by using an affine transformation.
  • the affine transformation may be implemented by a series of atomic transformations, which may include, but are not limited to, Translation, Scale, Flip, Rotation, and Miscut ( At least one of Shear).
  • the difference area image may be extracted from the to-be-identified image according to a pre-specified area location.
  • the location of the region with greater difference may be further pre-specified manually, as shown by the solid line frame in FIG. 3, various geometric shapes may be used to position the region. Calibration.
  • the upper left corner (x1, y1) is described as (x1/width, y1/height)
  • the lower right corner (x2, y2) is described as (x2/width, y2/height).
  • width, height is the image length of the template image and the image width of the template image.
  • the area may be specifically located from the image to be identified according to the pre-specified area position. Position, and further, the image covered by the area location may be extracted as the difference area image.
  • the image feature of the different area image may be obtained by using a model trained by the universal data set according to the difference area image.
  • template images of at least two specified categories may be further acquired, and further, templates of each specified category may be selected from at least two specified categories.
  • template features of each of the specified categories may be obtained according to the template area image of each specified category.
  • the template region image of each specified category may be extracted from the template image of each specified category according to a preset location of the region. Then, the template features of each of the specified categories may be obtained by using the obtained model according to the template area image.
  • the image feature of the difference area image and the preset template feature may be measured to obtain the closest template feature, and further, the template image to which the template feature belongs may be The category is used as the recognition result.
  • the image to be identified of a specified size is obtained, and then the image to be identified is Extracting a difference area image, and obtaining an image feature of the difference area image according to the difference area image, so that the image to be recognized can be obtained according to the image feature of the difference area image and a preset template feature.
  • the recognition result can realize the recognition processing of the image with a limited number of categories without using the deep learning method based on hundreds of thousands or even millions of training samples.
  • the training samples are trained to obtain a model by using a deep learning method, but the model trained by the universal data set can be used to remove The workload of data acquisition and model training for large-scale training samples can effectively speed up the algorithm development time of image recognition processing.
  • the accuracy of the image recognition processing can be effectively ensured by manually specifying the location of the region having a large difference.
  • FIG. 2 is a schematic structural diagram of an image recognition apparatus according to another embodiment of the present invention, as shown in FIG. 2 .
  • the image recognition apparatus of the present embodiment may include an acquisition unit 21, an extraction unit 22, a feature unit 23, and an identification unit 24.
  • the obtaining unit 21 is configured to acquire a specified size.
  • the image recognition apparatus may be an application located in a local terminal, or may be a plug-in or a software development kit (SDK) installed in an application located in the local terminal.
  • SDK software development kit
  • the unit may be a processing engine located in the network side server, or may be a distributed system located on the network side, which is not specifically limited in this embodiment.
  • the application may be a local application (nativeApp) installed on the terminal, or may be a web application (webApp) of the browser on the terminal, which is not specifically limited in this embodiment.
  • the acquiring unit 21 may be specifically configured to adjust, by using an affine transformation, the obtained image of any size to be identified to be the specified size to be identified. image.
  • the extracting unit 22 may be configured to extract the difference area image from the to-be-identified image according to a pre-specified area location.
  • the feature unit 23 may be specifically configured to obtain an image of the difference area image by using a model trained by the universal data set according to the difference area image. feature.
  • the feature unit 23 may be further configured to obtain template images of at least two specified categories; a template image for each of the specified categories in the category, extracting the template region image of each of the specified categories; and obtaining the template features of each of the specified categories according to the template region images of each of the specified categories.
  • the image to be identified of the specified size is acquired by the acquiring unit, and the difference area image is extracted from the image to be recognized by the extracting unit, and the image of the difference area is obtained by the feature unit according to the image of the difference area.
  • the image feature enables the recognition unit to obtain the recognition result of the image to be recognized according to the image feature of the difference region image and the preset template feature, without using deep learning based on hundreds of thousands or even millions of training samples In this way, it is possible to identify and process images with a limited number of categories.
  • the training samples are trained to obtain a model by using a deep learning method, but the model trained by the universal data set can be used to remove The workload of data acquisition and model training for large-scale training samples can effectively speed up the algorithm development time of image recognition processing.
  • the accuracy of the image recognition processing can be effectively ensured by manually specifying the location of the region having a large difference.
  • the disclosed system is The method of setting can be implemented in other ways.
  • the device embodiments described above are merely illustrative.
  • the division of the unit is only a logical function division.
  • there may be another division manner for example, multiple units 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 connection shown or discussed may be an indirect coupling or communication connection through some interface, device or unit, and may be in an electrical, mechanical or other form.
  • the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, may be located in one place, or may be distributed to multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of the embodiment.
  • each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.
  • the above integrated unit can be implemented in the form of hardware or in the form of hardware plus software functional units.
  • the above-described integrated unit implemented in the form of a software functional unit can be stored in a computer readable storage medium.
  • the above software functional unit is stored in a storage medium and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) or a processor to perform the methods of the various embodiments of the present invention. Part of the steps.
  • the foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, and the like, which can store program codes. .

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Data Mining & Analysis (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Databases & Information Systems (AREA)
  • Multimedia (AREA)
  • General Engineering & Computer Science (AREA)
  • Medical Informatics (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • Software Systems (AREA)
  • Library & Information Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Image Analysis (AREA)

Abstract

本发明提供一种图像识别方法、装置、设备及非易失性计算机存储介质。本发明实施例通过获取指定尺寸的待识别图像,进而从所述待识别图像中,提取差异区域图像,并根据所述差异区域图像,获得所述差异区域图像的图像特征,使得能够根据所述差异区域图像的图像特征和预先设置的模板特征,获得所述待识别图像的识别结果,无需基于几十万甚至上百万个训练样本采用深度学习的方法,就能够实现对类别数量有限的图像进行识别处理。

Description

图像识别方法、装置、设备及非易失性计算机存储介质
本申请要求了申请日为2016年03月14日,申请号为201610143523.0发明名称为“图像识别方法及装置”的中国专利申请的优先权。
技术领域
本发明涉及图像处理技术,尤其涉及一种图像识别方法、装置、设备及非易失性计算机存储介质。
背景技术
近些年来,采用深度学习的方法,在图像识别领域取得了很不错的结果。深度学习对于训练样本的数量要求较高,其数量往往在几十万甚至上百万个训练样本。
然而,对于图像的类别数量有限的情况,其训练样本的数量也是非常有限的,并不太适合采用深度学习的方法,对这些类别数量有限的图像进行识别处理。因此,亟需提供一种图像识别方法,对类别数量有限的图像进行识别处理。
发明内容
本发明的多个方面提供一种图像识别方法、装置、设备及非易失性计算机存储介质,用以对类别数量有限的图像进行识别处理。
本发明的一方面,提供一种图像识别方法,包括:
获取指定尺寸的待识别图像;
从所述待识别图像中,提取差异区域图像;
根据所述差异区域图像,获得所述差异区域图像的图像特征;
根据所述差异区域图像的图像特征和预先设置的模板特征,获得所述待识别图像的识别结果。
如上所述的方面和任一可能的实现方式,进一步提供一种实现方式,所述获取指定尺寸的待识别图像,包括:
利用仿射变换,将所获得的任意尺寸的待识别图像调整为所述指定尺寸的待识别图像。
如上所述的方面和任一可能的实现方式,进一步提供一种实现方式,所述从所述待识别图像中,提取差异区域图像,包括:
根据预先指定的区域位置,从所述待识别图像中,提取所述差异区域图像。
如上所述的方面和任一可能的实现方式,进一步提供一种实现方式,所述根据所述差异区域图像,获得所述差异区域图像的图像特征,包括:
根据所述差异区域图像,利用通用数据集合所训练的模型,获得所述差异区域图像的图像特征。
如上所述的方面和任一可能的实现方式,进一步提供一种实现方式, 所述根据所述差异区域图像的图像特征和预先设置的模板特征,获得所述待识别图像的识别结果之前,还包括:
获取至少两个指定类别的模板图像;
从至少两个指定类别中每个指定类别的模板图像,提取所述每个指定类别的模板区域图像;
根据所述每个指定类别的模板区域图像,获得所述每个指定类别的模板特征。
本发明的另一方面,提供一种图像识别装置,包括:
获取单元,用于获取指定尺寸的待识别图像;
提取单元,用于从所述待识别图像中,提取差异区域图像;
特征单元,用于根据所述差异区域图像,获得所述差异区域图像的图像特征;
识别单元,用于根据所述差异区域图像的图像特征和预先设置的模板特征,获得所述待识别图像的识别结果。
如上所述的方面和任一可能的实现方式,进一步提供一种实现方式,所述获取单元,具体用于
利用仿射变换,将所获得的任意尺寸的待识别图像调整为所述指定尺寸的待识别图像。
如上所述的方面和任一可能的实现方式,进一步提供一种实现方式,所述提取单元,具体用于
根据预先指定的区域位置,从所述待识别图像中,提取所述差异区域图像。
如上所述的方面和任一可能的实现方式,进一步提供一种实现方式,所述特征单元,具体用于
根据所述差异区域图像,利用通用数据集合所训练的模型,获得所述差异区域图像的图像特征。
如上所述的方面和任一可能的实现方式,进一步提供一种实现方式,所述特征单元,还用于
获取至少两个指定类别的模板图像;
从至少两个指定类别中每个指定类别的模板图像,提取所述每个指定类别的模板区域图像;以及
根据所述每个指定类别的模板区域图像,获得所述每个指定类别的模板特征。
本发明的另一方面,提供一种设备,包括:
一个或者多个处理器;
存储器;
一个或者多个程序,所述一个或者多个程序存储在所述存储器中,当被所述一个或者多个处理器执行时:
获取指定尺寸的待识别图像;
从所述待识别图像中,提取差异区域图像;
根据所述差异区域图像,获得所述差异区域图像的图像特征;
根据所述差异区域图像的图像特征和预先设置的模板特征,获得所述待识别图像的识别结果。
本发明的另一方面,提供一种非易失性计算机存储介质,所述非易失性计算机存储介质存储有一个或者多个程序,当所述一个或者多个程序被一个设备执行时,使得所述设备:
获取指定尺寸的待识别图像;
从所述待识别图像中,提取差异区域图像;
根据所述差异区域图像,获得所述差异区域图像的图像特征;
根据所述差异区域图像的图像特征和预先设置的模板特征,获得所述待识别图像的识别结果。
由上述技术方案可知,本发明实施例通过获取指定尺寸的待识别图像,进而从所述待识别图像中,提取差异区域图像,并根据所述差异区域图像,获得所述差异区域图像的图像特征,使得能够根据所述差异区域图像的图像特征和预先设置的模板特征,获得所述待识别图像的识别结果,无需基于几十万甚至上百万个训练样本采用深度学习的方法,就能够实现对类别数量有限的图像进行识别处理。
另外,采用本发明所提供的技术方案,不需要专门采集大规模的训练样本,采用深度学习的方法,对这些训练样本进行训练获得模型,而 是可以利用通用数据集合所训练的模型,使得去除了大规模训练样本的数据采集和模型训练的工作量,能够有效加快图像识别处理的算法开发时间。
另外,采用本发明所提供的技术方案,通过人工预先指定具有较大差异性的区域位置,能够有效保证图像识别处理的准确性。
附图说明
为了更清楚地说明本发明实施例中的技术方案,下面将对实施例或现有技术描述中所需要使用的附图作一简单地介绍,显而易见地,下面描述中的附图是本发明的一些实施例,对于本领域普通技术人员来讲,在不付出创造性劳动性的前提下,还可以根据这些附图获得其他的附图。
图1为本发明一实施例提供的图像识别方法的流程示意图;
图2为本发明另一实施例提供的图像识别装置的结构示意图;
图3为图1对应的实施例中差异区域图像的示意图。
具体实施方式
为使本发明实施例的目的、技术方案和优点更加清楚,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,所描述的实施例是本发明一部分实施例,而不是全部的实施例。基于本发明中的实施例,本领域普通技术人员在没有作出创造性劳动前提下所获得的全部其他实施例,都属于本发明保护的范围。
需要说明的是,本发明实施例中所涉及的终端可以包括但不限于手 机、个人数字助理(Personal Digital Assistant,PDA)、无线手持设备、平板电脑(Tablet Computer)、个人电脑(Personal Computer,PC)、MP3播放器、MP4播放器、可穿戴设备(例如,智能眼镜、智能手表、智能手环等)等。
另外,本文中术语“和/或”,仅仅是一种描述关联对象的关联关系,表示可以存在三种关系,例如,A和/或B,可以表示:单独存在A,同时存在A和B,单独存在B这三种情况。另外,本文中字符“/”,一般表示前后关联对象是一种“或”的关系。
图1为本发明一实施例提供的图像识别方法的流程示意图,如图1所示。
101、获取指定尺寸的待识别图像。
所谓的图像,是指采用一定的图像格式,将图像数据即图像的像素按照一定的方式进行存储,所形成的文件,又可以称为图像文件。
其中,图像的图像格式即图像存储的格式,可以包括但不限于位图(Bitmap,BMP)格式、可移植网络图像格式(Portable Network Graphic Format,PNG)、联合图像专家组(Joint Photographic Experts Group,JPEG)格式、可交换图像文件格式(Exchangeable Image File Format,EXIF),本实施例对此不进行特别限定。
102、从所述待识别图像中,提取差异区域图像。
103、根据所述差异区域图像,获得所述差异区域图像的图像特征。
104、根据所述差异区域图像的图像特征和预先设置的模板特征,获得所述待识别图像的识别结果。
需要说明的是,101~104的执行主体可以为位于本地终端的应用, 或者还可以为设置在位于本地终端的应用中的插件或软件开发工具包(Software Development Kit,SDK)等功能单元,或者还可以为位于网络侧服务器中的处理引擎,或者还可以为位于网络侧的分布式***,本实施例对此不进行特别限定。
可以理解的是,所述应用可以是安装在终端上的本地程序(nativeApp),或者还可以是终端上的浏览器的一个网页程序(webApp),本实施例对此不进行特别限定。
这样,通过获取指定尺寸的待识别图像,进而从所述待识别图像中,提取差异区域图像,并根据所述差异区域图像,获得所述差异区域图像的图像特征,使得能够根据所述差异区域图像的图像特征和预先设置的模板特征,获得所述待识别图像的识别结果,无需基于几十万甚至上百万个训练样本采用深度学习的方法,就能够实现对类别数量有限的图像进行识别处理。
本发明中,所述待识别图像可以为利用图像传感器,所采集的。其中,所述图像传感器可以为电荷耦合元件(Charge Coupled Device,CCD)传感器,或者还可以为金属氧化物半导体元件(Complementary Metal-Oxide Semiconductor,CMOS)传感器,本实施例对此不进行特别限定。
由于所采集的图像中除了包含待识别图像所对应的目标物体之外,通常还会包含一些其他物体作为背景图像,例如,一个人手里拿着一张人民币的图像中,除了包含待识别图像所对应的目标物体即人民币之外,还可能会包含人手、收银台等一些其他物体作为背景图像,等等,因此,还需要进一步采用传统的图像检测的方法,例如,尺度不变特征变换 (Scale-Invariant Feature Transform,SIFT)算法等,找到图像中的目标物体的区域,以作为所述待识别图像。
可选地,在本实施例的一个可能的实现方式中,在101中,具体可以利用仿射变换,将所获得的任意尺寸的待识别图像调整为所述指定尺寸的待识别图像。
具体来说,所述仿射变换可以通过一系列的原子变换的复合来实现,具体可以包括但不限于平移(Translation)、缩放(Scale)、翻转(Flip)、旋转(Rotation)和错切(Shear)中的至少一项。
可选地,在本实施例的一个可能的实现方式中,在102中,具体可以根据预先指定的区域位置,从所述待识别图像中,提取所述差异区域图像。
在一个具体的实现过程中,在102之前,可以进一步由人工预先指定具有较大差异性的区域位置,如图3中的实线框所示,可以采用各种几何形状来对该区域位置进行标定。
例如,可以一个矩形框的位置来标定,即左上角(x1,y1),右下角(x2,y2)。为了便于计算,可以采用比例来描述,将左上角(x1,y1)描述为(x1/width,y1/height),将右下角(x2,y2)描述为(x2/width,y2/height),其中width,height为模板图像的图像长度和模板图像的图像宽度。
这样,通过人工预先指定具有较大差异性的区域位置,能够有效保证图像识别处理的准确性。
在该实现方式中,在人工预先指定具有较大差异性的区域位置之后,具体可以根据预先指定的区域位置,从所述待识别图像中,定位该区域 位置,进而,则可以将该区域位置所覆盖的图像提取出来,以作为所述差异区域图像。
可选地,在本实施例的一个可能的实现方式中,在103中,具体可以根据所述差异区域图像,利用通用数据集合所训练的模型,获得所述差异区域图像的图像特征。
在该实现方式中,在103之前,还需要基于现有的通用数据集合,采用深度学习算法,进行样本训练,获得一个模型,例如,采用深层神经网络(Deep Neural Network,CDNN),基于ImageNet(图像识别目前最大的数据库)所公开的数据集合,训练所得到的模型,该模型一般很容易获取。
可选地,在本实施例的一个可能的实现方式中,在104之前,还可以进一步获取至少两个指定类别的模板图像,进而,则可以从至少两个指定类别中每个指定类别的模板图像,提取所述每个指定类别的模板区域图像。然后,则可以根据所述每个指定类别的模板区域图像,获得所述每个指定类别的模板特征。
在该实现方式中,具体可以根据预先指定的区域位置,从每个指定类别的模板图像,提取所述每个指定类别的模板区域图像。然后,则可以根据所述模板区域图像,利用所获得的所述模型,获得所述每个指定类别的模板特征。
具体来说,在104中,具体可以将对所述差异区域图像的图像特征和预先设置的模板特征进行度量距离,获得距离最近的模板特征,进而,则可以将该模板特征所属的模板图像的类别作为识别结果。
本实施例中,通过获取指定尺寸的待识别图像,进而从所述待识别 图像中,提取差异区域图像,并根据所述差异区域图像,获得所述差异区域图像的图像特征,使得能够根据所述差异区域图像的图像特征和预先设置的模板特征,获得所述待识别图像的识别结果,无需基于几十万甚至上百万个训练样本采用深度学习的方法,就能够实现对类别数量有限的图像进行识别处理。
另外,采用本发明所提供的技术方案,不需要专门采集大规模的训练样本,采用深度学习的方法,对这些训练样本进行训练获得模型,而是可以利用通用数据集合所训练的模型,使得去除了大规模训练样本的数据采集和模型训练的工作量,能够有效加快图像识别处理的算法开发时间。
另外,采用本发明所提供的技术方案,通过人工预先指定具有较大差异性的区域位置,能够有效保证图像识别处理的准确性。
需要说明的是,对于前述的各方法实施例,为了简单描述,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他顺序或者同时进行。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是本发明所必须的。
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,可以参见其他实施例的相关描述。
图2为本发明另一实施例提供的图像识别装置的结构示意图,如图2所示。本实施例的图像识别装置可以包括获取单元21、提取单元22、特征单元23和识别单元24。其中,获取单元21,用于获取指定尺寸的 待识别图像;提取单元22,用于从所述待识别图像中,提取差异区域图像;特征单元23,用于根据所述差异区域图像,获得所述差异区域图像的图像特征;识别单元24,用于根据所述差异区域图像的图像特征和预先设置的模板特征,获得所述待识别图像的识别结果。
需要说明的是,本实施例所提供的图像识别装置可以为位于本地终端的应用,或者还可以为设置在位于本地终端的应用中的插件或软件开发工具包(Software Development Kit,SDK)等功能单元,或者还可以为位于网络侧服务器中的处理引擎,或者还可以为位于网络侧的分布式***,本实施例对此不进行特别限定。
可以理解的是,所述应用可以是安装在终端上的本地程序(nativeApp),或者还可以是终端上的浏览器的一个网页程序(webApp),本实施例对此不进行特别限定。
可选地,在本实施例的一个可能的实现方式中,所述获取单元21,具体可以用于利用仿射变换,将所获得的任意尺寸的待识别图像调整为所述指定尺寸的待识别图像。
可选地,在本实施例的一个可能的实现方式中,所述提取单元22,具体可以用于根据预先指定的区域位置,从所述待识别图像中,提取所述差异区域图像。
可选地,在本实施例的一个可能的实现方式中,所述特征单元23,具体可以用于根据所述差异区域图像,利用通用数据集合所训练的模型,获得所述差异区域图像的图像特征。
可选地,在本实施例的一个可能的实现方式中,所述特征单元23,还可以进一步用于获取至少两个指定类别的模板图像;从至少两个指定 类别中每个指定类别的模板图像,提取所述每个指定类别的模板区域图像;以及根据所述每个指定类别的模板区域图像,获得所述每个指定类别的模板特征。
需要说明的是,图1对应的实施例中方法,可以由本实施例提供的图像识别装置实现。详细描述可以参见图1对应的实施例中的相关内容,此处不再赘述。
本实施例中,通过获取单元获取指定尺寸的待识别图像,进而由提取单元从所述待识别图像中,提取差异区域图像,并由特征单元根据所述差异区域图像,获得所述差异区域图像的图像特征,使得识别单元能够根据所述差异区域图像的图像特征和预先设置的模板特征,获得所述待识别图像的识别结果,无需基于几十万甚至上百万个训练样本采用深度学习的方法,就能够实现对类别数量有限的图像进行识别处理。
另外,采用本发明所提供的技术方案,不需要专门采集大规模的训练样本,采用深度学习的方法,对这些训练样本进行训练获得模型,而是可以利用通用数据集合所训练的模型,使得去除了大规模训练样本的数据采集和模型训练的工作量,能够有效加快图像识别处理的算法开发时间。
另外,采用本发明所提供的技术方案,通过人工预先指定具有较大差异性的区域位置,能够有效保证图像识别处理的准确性。
所属领域的技术人员可以清楚地了解到,为描述的方便和简洁,上述描述的***,装置和单元的具体工作过程,可以参考前述方法实施例中的对应过程,在此不再赘述。
在本发明所提供的几个实施例中,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,所述单元的划分,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,例如多个单元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连接可以是通过一些接口,装置或单元的间接耦合或通信连接,可以是电性,机械或其它的形式。
所述作为分离部件说明的单元可以是或者也可以不是物理上分开的,作为单元显示的部件可以是或者也可以不是物理单元,即可以位于一个地方,或者也可以分布到多个网络单元上。可以根据实际的需要选择其中的部分或者全部单元来实现本实施例方案的目的。
另外,在本发明各个实施例中的各功能单元可以集成在一个处理单元中,也可以是各个单元单独物理存在,也可以两个或两个以上单元集成在一个单元中。上述集成的单元既可以采用硬件的形式实现,也可以采用硬件加软件功能单元的形式实现。
上述以软件功能单元的形式实现的集成的单元,可以存储在一个计算机可读取存储介质中。上述软件功能单元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)或处理器(processor)执行本发明各个实施例所述方法的部分步骤。而前述的存储介质包括:U盘、移动硬盘、只读存储器(Read-Only Memory,ROM)、随机存取存储器(Random Access Memory,RAM)、磁碟或者光盘等各种可以存储程序代码的介质。
最后应说明的是:以上实施例仅用以说明本发明的技术方案,而非 对其限制;尽管参照前述实施例对本发明进行了详细的说明,本领域的普通技术人员应当理解:其依然可以对前述各实施例所记载的技术方案进行修改,或者对其中部分技术特征进行等同替换;而这些修改或者替换,并不使相应技术方案的本质脱离本发明各实施例技术方案的精神和范围。

Claims (12)

  1. 一种图像识别方法,其特征在于,包括:
    获取指定尺寸的待识别图像;
    从所述待识别图像中,提取差异区域图像;
    根据所述差异区域图像,获得所述差异区域图像的图像特征;
    根据所述差异区域图像的图像特征和预先设置的模板特征,获得所述待识别图像的识别结果。
  2. 根据权利要求1所述的方法,其特征在于,所述获取指定尺寸的待识别图像,包括:
    利用仿射变换,将所获得的任意尺寸的待识别图像调整为所述指定尺寸的待识别图像。
  3. 根据权利要求1或2所述的方法,其特征在于,所述从所述待识别图像中,提取差异区域图像,包括:
    根据预先指定的区域位置,从所述待识别图像中,提取所述差异区域图像。
  4. 根据权利要求1~3任一权利要求所述的方法,其特征在于,所述根据所述差异区域图像,获得所述差异区域图像的图像特征,包括:
    根据所述差异区域图像,利用通用数据集合所训练的模型,获得所述差异区域图像的图像特征。
  5. 根据权利要求1~4任一权利要求所述的方法,其特征在于,所述根据所述差异区域图像的图像特征和预先设置的模板特征,获得所述待识别图像的识别结果之前,还包括:
    获取至少两个指定类别的模板图像;
    从至少两个指定类别中每个指定类别的模板图像,提取所述每个指定类别的模板区域图像;
    根据所述每个指定类别的模板区域图像,获得所述每个指定类别的模板特征。
  6. 一种图像识别装置,其特征在于,包括:
    获取单元,用于获取指定尺寸的待识别图像;
    提取单元,用于从所述待识别图像中,提取差异区域图像;
    特征单元,用于根据所述差异区域图像,获得所述差异区域图像的图像特征;
    识别单元,用于根据所述差异区域图像的图像特征和预先设置的模板特征,获得所述待识别图像的识别结果。
  7. 根据权利要求6所述的装置,其特征在于,所述获取单元,具体用于
    利用仿射变换,将所获得的任意尺寸的待识别图像调整为所述指定尺寸的待识别图像。
  8. 根据权利要求6或7所述的装置,其特征在于,所述提取单元,具体用于
    根据预先指定的区域位置,从所述待识别图像中,提取所述差异区域图像。
  9. 根据权利要求6~8任一权利要求所述的装置,其特征在于,所述特征单元,具体用于
    根据所述差异区域图像,利用通用数据集合所训练的模型,获得所述差异区域图像的图像特征。
  10. 根据权利要求6~9任一权利要求所述的装置,其特征在于,所述特征单元,还用于
    获取至少两个指定类别的模板图像;
    从至少两个指定类别中每个指定类别的模板图像,提取所述每个指定类别的模板区域图像;以及
    根据所述每个指定类别的模板区域图像,获得所述每个指定类别的模板特征。
  11. 一种设备,包括:
    一个或者多个处理器;
    存储器;
    一个或者多个程序,所述一个或者多个程序存储在所述存储器中,当被所述一个或者多个处理器执行时:
    获取指定尺寸的待识别图像;
    从所述待识别图像中,提取差异区域图像;
    根据所述差异区域图像,获得所述差异区域图像的图像特征;
    根据所述差异区域图像的图像特征和预先设置的模板特征,获得所述待识别图像的识别结果。
  12. 一种非易失性计算机存储介质,所述非易失性计算机存储介质存储有一个或者多个程序,当所述一个或者多个程序被一个设备执行时,使得所述设备:
    获取指定尺寸的待识别图像;
    从所述待识别图像中,提取差异区域图像;
    根据所述差异区域图像,获得所述差异区域图像的图像特征;
    根据所述差异区域图像的图像特征和预先设置的模板特征,获得所述待识别图像的识别结果。
PCT/CN2016/082969 2016-03-14 2016-05-23 图像识别方法、装置、设备及非易失性计算机存储介质 WO2017156864A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US16/080,945 US11455783B2 (en) 2016-03-14 2016-05-23 Image recognition method and apparatus, device and non-volatile computer storage medium

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201610143523.0 2016-03-14
CN201610143523.0A CN105825228B (zh) 2016-03-14 2016-03-14 图像识别方法及装置

Publications (1)

Publication Number Publication Date
WO2017156864A1 true WO2017156864A1 (zh) 2017-09-21

Family

ID=56988099

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2016/082969 WO2017156864A1 (zh) 2016-03-14 2016-05-23 图像识别方法、装置、设备及非易失性计算机存储介质

Country Status (3)

Country Link
US (1) US11455783B2 (zh)
CN (1) CN105825228B (zh)
WO (1) WO2017156864A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113239804A (zh) * 2021-05-13 2021-08-10 杭州睿胜软件有限公司 图像识别方法、可读存储介质及图像识别***

Families Citing this family (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108734185B (zh) * 2017-04-18 2022-01-04 北京京东尚科信息技术有限公司 图像校验方法和装置
JP7326867B2 (ja) * 2019-05-21 2023-08-16 富士通株式会社 情報処理装置、情報処理プログラム及び情報処理方法
CN110378876A (zh) * 2019-06-18 2019-10-25 平安科技(深圳)有限公司 基于深度学习的影像识别方法、装置、设备及存储介质
CN112330552A (zh) * 2020-10-21 2021-02-05 重庆锐云科技有限公司 图片处理方法、服务器及计算机可读存储介质

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101299239A (zh) * 2008-06-06 2008-11-05 北京中星微电子有限公司 文字区域图像的获取方法及装置、文字识别***
CN103927511A (zh) * 2014-02-25 2014-07-16 华北电力大学(保定) 基于差异特征描述的图像识别方法
CN104636726A (zh) * 2015-02-05 2015-05-20 深圳市中兴移动通信有限公司 一种图像色彩识别方法、装置及终端
CN104751108A (zh) * 2013-12-31 2015-07-01 汉王科技股份有限公司 人脸图像识别装置和人脸图像识别方法

Family Cites Families (14)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7607016B2 (en) * 2001-04-20 2009-10-20 Digimarc Corporation Including a metric in a digital watermark for media authentication
US20040213448A1 (en) * 2003-04-28 2004-10-28 Asn Technology Corp. Apparatus for recognizing counterfeit currency and method thereof
DE602006009839D1 (de) * 2005-12-05 2009-11-26 Siemens Corp Res Inc Verwendung von sequentiellem clustering zur instanzenauswahl bei der maschinenzustandsüberwachung
KR101487944B1 (ko) * 2010-02-24 2015-01-30 아이피플렉 홀딩스 코포레이션 시각 장애인들을 지원하는 증강 현실 파노라마
CN102375987B (zh) * 2010-08-17 2014-04-02 国基电子(上海)有限公司 图像处理设备及图像特征向量提取与图像匹配方法
AU2011302414A1 (en) * 2010-09-15 2013-05-02 Identicoin, Inc. Coin identification method and apparatus
CN102486868A (zh) * 2010-12-06 2012-06-06 华南理工大学 一种基于平均脸的美丽人脸合成方法
US9351641B2 (en) * 2012-10-04 2016-05-31 Cerner Innovation, Inc. Mobile processing device system for patient monitoring data acquisition
CN103810488B (zh) * 2012-11-09 2017-08-08 阿里巴巴集团控股有限公司 图像特征提取方法、图像搜索方法、服务器、终端及***
US9378604B1 (en) * 2013-06-27 2016-06-28 Tim E. Rathjen Coin recognition system and method
US10169686B2 (en) * 2013-08-05 2019-01-01 Facebook, Inc. Systems and methods for image classification by correlating contextual cues with images
US9500595B2 (en) * 2014-12-26 2016-11-22 M.A.C.Ge, Llc Methods for enhanced grading of mint quality coins
WO2016148247A1 (ja) * 2015-03-18 2016-09-22 国立研究開発法人理化学研究所 球状体の回転の測定装置、測定方法、ならびに、プログラム
CN105320945A (zh) * 2015-10-30 2016-02-10 小米科技有限责任公司 图像分类的方法及装置

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101299239A (zh) * 2008-06-06 2008-11-05 北京中星微电子有限公司 文字区域图像的获取方法及装置、文字识别***
CN104751108A (zh) * 2013-12-31 2015-07-01 汉王科技股份有限公司 人脸图像识别装置和人脸图像识别方法
CN103927511A (zh) * 2014-02-25 2014-07-16 华北电力大学(保定) 基于差异特征描述的图像识别方法
CN104636726A (zh) * 2015-02-05 2015-05-20 深圳市中兴移动通信有限公司 一种图像色彩识别方法、装置及终端

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113239804A (zh) * 2021-05-13 2021-08-10 杭州睿胜软件有限公司 图像识别方法、可读存储介质及图像识别***
CN113239804B (zh) * 2021-05-13 2023-06-02 杭州睿胜软件有限公司 图像识别方法、可读存储介质及图像识别***

Also Published As

Publication number Publication date
CN105825228A (zh) 2016-08-03
US11455783B2 (en) 2022-09-27
US20190080195A1 (en) 2019-03-14
CN105825228B (zh) 2019-04-30

Similar Documents

Publication Publication Date Title
WO2017156864A1 (zh) 图像识别方法、装置、设备及非易失性计算机存储介质
WO2020125216A1 (zh) 一种行人重识别方法、装置、电子设备及计算机可读存储介质
WO2019109526A1 (zh) 人脸图像的年龄识别方法、装置及存储介质
US10534957B2 (en) Eyeball movement analysis method and device, and storage medium
US11270099B2 (en) Method and apparatus for generating facial feature
JP6030240B2 (ja) 顔認識のための方法および装置
CN110427859A (zh) 一种人脸检测方法、装置、电子设备及存储介质
US9846808B2 (en) Image integration search based on human visual pathway model
JP6624794B2 (ja) 画像処理装置、画像処理方法及びプログラム
WO2019033571A1 (zh) 面部特征点检测方法、装置及存储介质
WO2019061658A1 (zh) 眼镜定位方法、装置及存储介质
US11145080B2 (en) Method and apparatus for three-dimensional object pose estimation, device and storage medium
US10650234B2 (en) Eyeball movement capturing method and device, and storage medium
Vazquez-Fernandez et al. Built-in face recognition for smart photo sharing in mobile devices
WO2016106966A1 (zh) 人物标注方法和终端、存储介质
WO2022174605A1 (zh) 一种手势识别方法、手势识别装置及智能设备
KR101559502B1 (ko) 실시간 손 포즈 인식을 통한 비접촉식 입력 인터페이스 방법 및 기록 매체
WO2019127102A1 (zh) 信息处理方法、装置、云处理设备以及计算机程序产品
WO2019033568A1 (zh) 嘴唇动作捕捉方法、装置及存储介质
CN112820071A (zh) 一种行为识别方法和装置
US20160110909A1 (en) Method and apparatus for creating texture map and method of creating database
JP2013037539A (ja) 画像特徴量抽出装置およびそのプログラム
CN103744903B (zh) 一种基于草图的场景图像检索方法
CN109711287B (zh) 人脸采集方法及相关产品
WO2024077785A1 (zh) 基于卷积神经网络模型的图像识别方法、装置及终端设备

Legal Events

Date Code Title Description
NENP Non-entry into the national phase

Ref country code: DE

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

Ref document number: 16894026

Country of ref document: EP

Kind code of ref document: A1

122 Ep: pct application non-entry in european phase

Ref document number: 16894026

Country of ref document: EP

Kind code of ref document: A1