WO2016107103A1 - 图像主体区域的识别方法及装置 - Google Patents
图像主体区域的识别方法及装置 Download PDFInfo
- Publication number
- WO2016107103A1 WO2016107103A1 PCT/CN2015/082049 CN2015082049W WO2016107103A1 WO 2016107103 A1 WO2016107103 A1 WO 2016107103A1 CN 2015082049 W CN2015082049 W CN 2015082049W WO 2016107103 A1 WO2016107103 A1 WO 2016107103A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- target candidate
- image
- body region
- subject
- regions
- Prior art date
- Legal status (The legal status 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 status listed.)
- Ceased
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/51—Indexing; Data structures therefor; Storage structures
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F16/583—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
- G06F16/5838—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content using colour
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
- G06V10/75—Organisation of the matching processes, e.g. simultaneous or sequential comparisons of image or video features; Coarse-fine approaches, e.g. multi-scale approaches; using context analysis; Selection of dictionaries
- G06V10/757—Matching configurations of points or features
Definitions
- the present invention relates to the field of computer technologies, and in particular, to a method and an apparatus for identifying an image body area.
- the extraction of image information is mainly concentrated on three levels of granularity, one is to classify coarse-grained images in units of images as a whole, and the other is to obtain physical objects in images (such as pedestrians, faces, cars, etc.).
- the target detection method which requires training of specialized types of detectors for detection.
- the third is fine-grained image segmentation or parsing of image pixels. Image segmentation or profiling can perform pixel-level granular description of image body details.
- the image processing methods of the above three granular levels have problems in image classification.
- the process of processing the image is: detecting the body region where the subject is located from the image, and then performing subsequent analysis on the subject based on the body region, for example, classifying or identifying the subject. Obtaining the main body region in the image in this whole process is the key to obtain accurate image analysis results.
- the first method is the subject detection method based on the saliency region, which is remarkable.
- the performance quantity function processes the image to generate an energy distribution map of the image, and obtains a region with higher significance according to the energy distribution map, and uses the region as the main body region of the image.
- the second way is the subject coordinate regression based on deep learning, which obtains the image to be processed, and then predicts the four point coordinates of the circumscribed rectangle of the body through the trained deep neural network to realize the detection and positioning of the body region.
- the problems in the above two methods are: the saliency energy function used in the first method is not universal, the accuracy and effectiveness of the detection result of the main body region are insufficient, and the second method not only calculates complicated and time-sensitive effects. Not good, but And this method can only return one body area, it is difficult to process images containing multiple subjects.
- the present invention aims to solve at least one of the technical problems in the related art to some extent.
- the first aspect of the present invention provides a method for recognizing an image body region, which can accurately obtain a body region of an image, and can identify a body corresponding to multiple bodies from an image including a plurality of subjects. region.
- An embodiment of the second aspect of the present invention is to provide an apparatus for identifying an image body area.
- a third aspect of the invention is directed to a storage medium.
- An embodiment of the fourth aspect of the present invention is to provide an identification device for an image body area.
- a method for recognizing an image body region includes: receiving an image to be recognized, and extracting a candidate body region from the image; and selecting a non-selection in the extracted candidate body region The subject area is excluded to obtain a target candidate subject area, and the target candidate subject area is adjusted and classified to obtain at least one target candidate subject area set; and a target candidate subject area in each target candidate subject area set Combine to identify the body area of the image.
- a candidate body region is first extracted from an image to be identified, and then the non-body region in the extracted candidate body region is excluded to obtain a target candidate body region, and The target candidate body region is adjusted and classified to obtain at least one target candidate body region set, and finally the target candidate body region in each target candidate body region set is merged to identify the body region of the image, thereby being accurate
- the body region of the image is obtained, and the body region corresponding to the plurality of bodies can be identified from the image including the plurality of subjects.
- an apparatus for identifying an image body region includes: an extraction module, configured to receive an image to be recognized, and extract a candidate body region from the image; and a processing module, configured to Excluding the non-subject regions in the extracted candidate subject regions to obtain target candidate subject regions, and adjusting and classifying the target candidate subject regions to obtain at least one set of target candidate subject region sets; and a merging module, A target candidate body region in each target candidate body region set is merged to identify a body region of the image.
- the candidate body region is extracted from the image to be identified by the extraction module, and then the non-body region in the extracted candidate body region is excluded by the processing module to obtain a target candidate. a body region, and adjusting and classifying the target candidate body region to obtain at least one target candidate body region set, and merging the target candidate body regions in each target candidate body region set to identify the body region of the image, Thereby, the body region of the image can be accurately obtained, and the body region corresponding to the plurality of subjects can be identified from the image including the plurality of subjects.
- a storage medium for storing an application, the application.
- the program is for performing the method of identifying an image body region according to the first aspect of the present invention.
- an identification device for an image body area includes: one or more processors; a memory; one or more modules, wherein the one or more modules are stored in the memory. And performing, by the one or more processors, an operation of: receiving an image to be recognized, and extracting a candidate body region from the image; and excluding the non-subject region in the extracted candidate body region, Obtaining a target candidate body region, and adjusting and classifying the target candidate body region to obtain at least one target candidate body region set; and merging target candidate body regions in each target candidate body region set to identify The body area of the image.
- FIG. 1 is a flow chart of a method of identifying an image body area in accordance with one embodiment of the present invention.
- FIG. 2 is a flow chart of a method of identifying an image body area in accordance with one embodiment of the present invention.
- FIG. 3 is an exemplary diagram of an unprocessed original image.
- FIG. 4 is a diagram showing an example of a result after extracting a candidate subject region in FIG. 3.
- FIG. 5 is a diagram showing an example of a result after the non-subject area is excluded from the candidate body region in FIG. 4.
- FIG. 6 is a diagram showing an example of a result of adjusting and classifying the candidate subject regions in FIG. 5.
- FIG. 7 is a diagram showing an example of a result of combining the target candidate body regions in FIG. 6.
- FIG. 8 is a block diagram showing the structure of an image body area identifying apparatus according to an embodiment of the present invention.
- FIG. 1 is a flowchart of a method for recognizing an image body region according to an embodiment of the present invention. As shown in FIG. 1, the method for identifying an image body region includes:
- S101 Receive an image to be identified, and extract a candidate body region from the image.
- the candidate body regions of different sizes and aspect ratios may be extracted from the image to be identified according to the general body detection model, so that the candidate body regions may cover various categories, different sizes, and different images in the image to be identified.
- the body content of the size usually a common body area detection model, can obtain a large number of candidate body areas of the image, which are usually indicated by rectangular boxes in the image.
- the universal subject detection model is a model for extracting candidate body regions of an image, and the model can be based on preset image features (eg, pixel grayscale, color, image gradient, edge, direction gradient histogram HOG (Histogram of Oriented Gradient), local binary pattern LBP (Local Binary Pattern), etc.) extracts the features of the image. After extracting the image to be recognized, it can pass the machine learning algorithm of the image (for example, support vector machine svm (support vector machine)
- the Adaboost algorithm (the algorithm is an adaptive integrated learning algorithm), a random forest algorithm, etc., analyzes the obtained features to obtain a candidate subject region corresponding to the image to be identified. It should be noted that the model extracts the candidate subject region of the image with high efficiency. Generally, the time required to extract the candidate subject region of one image is 100 ms.
- S102 Exclude the non-subject regions in the extracted candidate subject regions to obtain target candidate subject regions, and adjust and classify the target candidate subject regions to obtain at least one target candidate subject region set.
- the extracted candidate subject regions may be identified using different levels of classifiers to exclude non-subject regions in the candidate subject regions.
- the obtained candidate body regions may be identified by cascaded different levels of classifiers, and the candidate body regions are gradually excluded.
- different levels of classifiers have different decision criteria, and it is highly probable that the candidate body regions of all levels of classifiers become the body regions of the image.
- the obtained candidate body regions may be input into the first-level deep learning classifier, and the first-level deep learning classifier is based on The corresponding judging side identifies the candidate subject area, excludes the non-subject area that does not satisfy the subject area characteristic, and inputs the obtained recognition result into the second level deep learning classifier, and the second level deep learning classifier is based on the corresponding Judging the quasi-side and excluding some non-subject areas that do not satisfy the characteristics of the main body area, and outputting corresponding recognition results.
- the identification obtained by the second-level deep learning classifier may be obtained.
- the result is then input to a third-order linear classifier, which may also be referred to as a Support Vector Machine (SVM), and the third-level linear classifier identifies the input based on its corresponding judgment.
- SVM Support Vector Machine
- the result is further identified, and the non-subject area that does not satisfy the judgment criterion is removed, and the corresponding identification is output.
- the recognition result is output through said body region of all three candidate classifier.
- a large number of candidate subject areas can be coarsely screened in the early stage by using a low-level classifier, and in the latter stage, in order to accurately exclude a small number of indistinguishable non-subject areas in the set of candidate subject areas, A high-level classifier is used for identification to obtain an accurate corresponding target candidate body region.
- the candidate subject area passing through the classifiers of all levels is the target candidate subject area.
- the current target candidate region may be scaled according to the size relationship and the positional relationship between the subject and the current target candidate body region for each target candidate body region. Processing and/or translation processing.
- the scores of each target candidate body region before and after adjustment and the category to which it belongs can be obtained, and the corresponding target candidate body is matched according to the score of the target candidate subject region in the same category.
- the regions are sorted, and a target candidate body region set is obtained according to the sorting result.
- the obtained pre-adjusted and adjusted target candidate subject regions may be passed through a depth learning model-based classifier that classifies each target candidate subject region and outputs each target candidate subject region to each category.
- the probability value assuming that there are four target candidate subject areas before and after adjustment, the four target candidate subject areas are represented by a, b, c, and d, respectively, and the current classifier has three categories, respectively, using l, m and n represent that the probability distributions of the three categories l, m, and n corresponding to each target candidate body region obtained by the classifier are: a is (0.1, 0.1, 0.8), and b is (0.2, 0.2, 0.6) , c is (0.7, 0.1, 0.2), and d is (0.1, 0.85, 0.05).
- the obtained result is: a and b belong to the category n, and the probability of a is high (confidence high), before b, c Belong to category l, d belongs to category m.
- the target candidate body region may be merged according to the size relationship and the position relationship between the target candidate body regions.
- the target candidate subject regions may be merged according to the size relationship and the positional relationship between the target candidate subject regions, specifically, for the same
- the subject frame clustering algorithm may also be used, based on A machine learning algorithm such as a subject frame classification algorithm of a CRF (Conditional Random Fields) model combines target candidate body regions to obtain a body region corresponding to the two categories of subjects in the image.
- the image body region detecting manner of the embodiment can accurately obtain the body region of the corresponding body in the image, and can obtain the positional relationship of the body region in the image, which facilitates subsequent classification, recognition, and cropping of the image based on the obtained body region. Search, etc.
- the image can be cropped according to the body area of the image, and the cropped image is applied to the search result page to display the search result in the search result page.
- the method for identifying an image body region implemented by the present invention first extracts a candidate subject region from the image to be identified, and then excludes the non-subject region in the extracted candidate subject region to obtain a target candidate subject region, and selects a target candidate region.
- the main body area is adjusted and classified to obtain at least one target candidate main body area set, and finally the target candidate main body area in each target candidate main body area set is merged to identify the main body area of the image, thereby being
- the body region of the image is surely obtained, and the body region corresponding to the plurality of bodies can be identified from the image including the plurality of subjects.
- FIG. 2 is a flowchart of a method for recognizing an image body region according to another embodiment of the present invention.
- This embodiment uses a picture shown in FIG. 3 as an example to describe a process of recognizing an image body region in detail, as shown in FIG. 2 .
- the identification method of the main body area includes:
- the candidate body regions of different sizes and aspect ratios can be extracted from FIG. 3 according to the general body detection model.
- the results after extracting the candidate body regions in FIG. 3 are as shown in FIG. 4, and the rectangular frame in FIG. Representing a candidate subject area, each of the rectangular boxes of FIG. 4 represents one candidate subject area.
- the universal subject detection model is a model for extracting a candidate subject region of an image, and the model extracting the candidate subject region of the image is highly efficient. Generally, the time required to extract the candidate subject region of an image is 100 ms.
- a large number of candidate body regions are included in the candidate body region.
- a large number of extracted candidate body region sets can be identified by cascaded different classifiers.
- the non-subject area in the candidate subject area is gradually excluded to obtain the target candidate subject area.
- the possibility that the candidate body region of all levels of the classifier becomes the body region of the image is relatively large, and the candidate body region that passes through the classifiers of all levels may be referred to as the target candidate body region.
- FIG. 4 it can be seen from FIG. 4 that the figure includes two bodies, and therefore, the non-selective body regions corresponding to the two subjects can be non-classified by the multi-level two-class classifier.
- the main body area is excluded.
- the candidate main body area of FIG. 4 excludes the non-subject area, the obtained result is as shown in FIG. 5.
- FIG. 5 only includes several main body areas which may become images.
- Candidate body area ie target candidate body area).
- the current relationship between the body and the current target candidate body region may be
- the target candidate body region performs scaling processing and/or translation processing.
- the target candidate body region in FIG. 5 can be performed. Adjustment, in particular, for each target candidate body region in FIG. 5, zooming, panning, etc. may be performed according to the size relationship and the position key between the target candidate body region and the body to generate a new target candidate body region.
- the scores and categories of each target candidate body region before and after adjustment can be obtained, and the corresponding target candidate body regions are entered according to the scores of the target candidate body regions in the same category.
- the rows are sorted, and a target candidate body region collection is obtained according to the sorting result.
- the target candidate body region in FIG. 5 is adjusted, and then the pre-adjusted and adjusted candidate body regions are processed by a classifier, for example, a classifier based on a deep learning model, and the obtained result is as shown in FIG. 6.
- FIG. 6 shows a plurality of target candidate subject regions obtained.
- S204 Combine the target candidate body regions in each target candidate body region set to identify the body region of the image.
- the target candidate body regions are merged.
- the target candidate body region set may merge the target candidate body regions according to the size relationship and the position relationship between the target candidate body regions. Specifically, in the process of merging the target body regions in FIG. 6 , for the different subjects, the corresponding target candidate body regions around the body may be separately combined, specifically, according to the target candidate body regions. The size relationship and the positional relationship are merged. If the two target candidate body regions overlap, the circumscribed rectangles of the two can be merged. After the target candidate body regions in FIG. 6 are merged, the obtained body region of the image is as shown in the figure. As shown in Fig. 7, the two rectangular frames shown in Fig. 7 are the body regions of Fig. 3.
- the picture After obtaining the body area of FIG. 3, the picture may be described, classified, identified, cropped, etc. according to the obtained body area.
- the image may be cropped according to the body area of the image based on the obtained body area. And apply the cropped image to the search results page to display the search results graphically on the search results page.
- the method for identifying an image body region implemented by the present invention first extracts a candidate subject region from the image to be identified, and then excludes the non-subject region in the extracted candidate subject region to obtain a target candidate subject region, and selects a target candidate region.
- the body region is adjusted and classified to obtain at least one target candidate body region set, and finally the target candidate body region in each target candidate body region set is merged to identify the body region of the image, thereby being accurately obtained
- the body area of the image, and the body area corresponding to the plurality of bodies may be identified from the image including the plurality of bodies.
- the present invention also proposes an identification device for an image body area.
- FIG. 8 is a block diagram showing the structure of an image body area identifying apparatus according to an embodiment of the present invention.
- the identification device of the image body area includes: an extraction module 100, a processing module 200, and a merging module 300, wherein:
- the extraction module 100 is configured to receive an image to be identified, and extract a candidate body region set from the image; the processing module 200 For excluding the non-subject regions in the extracted candidate subject regions to obtain target candidate subject regions, and adjusting and classifying the target candidate subject regions to obtain at least one target candidate subject region set; and merging module 300 A target candidate body region in each target candidate body region set is merged to identify a body region of the image.
- the processing module 200 is specifically configured to: identify, for each candidate body region in the candidate body region set, the current candidate body region by using different levels of classifiers to exclude the non-subject region.
- the processing module 200 may identify the current candidate body region by cascading different levels of classifiers to exclude the non-body region in the candidate body region to obtain the target candidate body region.
- the processing module 200 may perform scaling processing on the current target candidate body region according to the size relationship and the position relationship between the body and the current target candidate body region. Or panning.
- the processing module 200 can obtain the score of each target candidate subject region before and after the adjustment and the category to which it belongs, and corresponding to the score pair of the target candidate subject region in the same category.
- the target candidate body regions are sorted, and a target candidate body region set is obtained according to the sorting result.
- the merging module 300 may merge the target candidate body regions according to the size relationship and the position relationship between the target candidate body regions.
- the process of identifying the image body area by the identification device of the image body area including the extraction module 100, the processing module 200, and the merging module 300 can be referred to FIG. 1 or FIG. 2 and corresponding text descriptions, and details are not described herein.
- the apparatus for identifying an image body region implemented by the present invention extracts a candidate body region from the image to be identified through an extraction module, and then excludes the non-body region in the extracted candidate body region by the processing module to obtain a target candidate body region. And adjusting and classifying the target candidate body region to obtain at least one target candidate body region set, and merging the target candidate body regions in each target candidate body region set to identify the body region of the image, thereby The body region of the image can be accurately obtained, and the body region corresponding to the plurality of bodies can be identified from the image including the plurality of subjects.
- the present invention also provides a storage medium for storing an application for performing an image body area identification method according to any of the embodiments of the present invention.
- the present invention also provides an identification device for an image body area, comprising: one or more processors; a memory; one or more modules, wherein the one or more modules are stored in the memory, When executed by the one or more processors, the following operations are performed:
- the target candidate body regions are adjusted and classified to obtain at least one target candidate body region set.
- first and second are used for descriptive purposes only and are not to be construed as indicating or implying a relative importance or implicitly indicating the number of technical features indicated.
- features defining “first” or “second” may include at least one of the features, either explicitly or implicitly.
- the meaning of "a plurality” is at least two, such as two, three, etc., unless specifically defined otherwise.
- a "computer-readable medium” can be any apparatus that can contain, store, communicate, propagate, or transport a program for use in an instruction execution system, apparatus, or device, or in conjunction with the instruction execution system, apparatus, or device.
- computer readable media include the following: electrical connections (electronic devices) having one or more wires, portable computer disk cartridges (magnetic devices), random access memory (RAM), Read only memory (ROM), erasable editable read only memory (EPROM or flash memory), fiber optic devices, and portable compact disk read only memory (CDROM).
- the computer readable medium may even be a paper or other suitable medium on which the program can be printed, as it may be optically scanned, for example by paper or other medium, followed by editing, interpretation or, if appropriate, other suitable The method is processed to obtain the program electronically and then stored in computer memory.
- portions of the invention may be implemented in hardware, software, firmware or a combination thereof.
- multiple steps or methods may be implemented in software or firmware stored in a memory and executed by a suitable instruction execution system to realise.
- a suitable instruction execution system to realise.
- if implemented in hardware as in another embodiment, it can be implemented by any one or combination of the following techniques well known in the art: having logic gates for implementing logic functions on data signals.
- Discrete logic circuits application specific integrated circuits with suitable combinational logic gates, programmable gate arrays (PGAs), field programmable gate arrays (FPGAs), etc.
- each functional unit in each embodiment of the present invention may be integrated into one processing module, or each unit may exist physically separately, or two or more units may be integrated into one module.
- the above integrated modules can be implemented in the form of hardware or in the form of software functional modules.
- the integrated modules, if implemented in the form of software functional modules and sold or used as stand-alone products, may also be stored in a computer readable storage medium.
- the above mentioned storage medium may be a read only memory, a magnetic disk or an optical disk or the like.
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- Library & Information Science (AREA)
- General Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Software Systems (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Computational Biology (AREA)
- Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Image Analysis (AREA)
Abstract
Description
Claims (12)
- 一种图像主体区域的识别方法,其特征在于,包括:接收待识别的图像,并从所述图像中提取候选主体区域;对所提取到的候选主体区域中的非主体区域进行排除,以获得目标候选主体区域,并对所述目标候选主体区域进行调整和分类,以获得至少一个目标候选主体区域集合;以及对每个目标候选主体区域集合中的目标候选主体区域合并,以识别出所述图像的主体区域。
- 根据权利要求1所述的方法,其特征在于,所述对所提取到的候选主体区域中的非主体区域进行排除,包括:使用不同级别的分类器对所提取到的候选主体区域进行识别,以排除所述非主体区域。
- 根据权利要求1或2所述的方法,其特征在于,所述对所述目标候选主体区域进行调整,包括:针对每个目标候选主体区域,根据主体和当前目标候选主体区域之间的大小关系和位置关系对所述当前目标候选主体区域进行缩放处理和/或平移处理。
- 根据权利要求1-3任一项所述的方法,其特征在于,所述对所述目标候选主体区域进行分类,包括:获得调整前和调整后的每个目标候选主体区域的评分和所属的类别,并根据同一类别中目标候选主体区域的评分对对应的目标候选主体区域进行排序,并根据排序结果获得所述目标候选主体区域集合。
- 根据权利要求1-4任一项所述的方法,其特征在于,所述对每个目标候选主体区域集合中的目标候选主体区域进行合并,以识别出所述图像的主体区域,包括:根据目标候选主体区域之间的大小关系及位置关系对目标候选主体区域进行合并。
- 一种图像主体区域的识别装置,其特征在于,包括:提取模块,用于接收待识别的图像,并从所述图像中提取候选主体区域;处理模块,用于对所提取到的候选主体区域中的非主体区域进行排除,以获得目标候选主体区域,并对所述目标候选主体区域进行调整和分类,以获得至少一个目标候选主体区域集合;以及合并模块,用于对每个目标候选主体区域集合中的目标候选主体区域合并,以识别出所述图像的主体区域。
- 根据权利要求6所述的装置,其特征在于,所述处理模块,具体用于:使用不同级别的分类器对所提取到的候选主体区域进行识别,以排除所述非主体区域。
- 根据权利要求6或7所述的装置,其特征在于,所述处理模块,具体用于:针对每个目标候选主体区域,根据主体和当前目标候选主体区域之间的大小关系和位置关系对所述当前目标候选主体区域进行缩放处理和/或平移处理。
- 根据权利要求6-8任一项所述的装置,其特征在于,所述处理模块,还用于:获得调整前和调整后的每个目标候选主体区域的评分和所属的类别,并根据同一类别中目标候选主体区域的评分对对应的目标候选主体区域进行排序,并根据排序结果获得所述目标候选主体区域集合。
- 根据权利要求6-9任一项所述的装置,其特征在于,所述合并模块,具体用于:根据目标候选主体区域之间的大小关系及位置关系对目标候选主体区域进行合并。
- 一种存储介质,其特征在于,用于存储应用程序,所述应用程序用于执行权利要求1至5中任一项所述的图像主体区域的识别方法。
- 一种图像主体区域的识别设备,其特征在于,包括:一个或者多个处理器;存储器;一个或者多个模块,所述一个或者多个模块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进行如下操作:接收待识别的图像,并从所述图像中提取候选主体区域;对所提取到的候选主体区域中的非主体区域进行排除,以获得目标候选主体区域,并对所述目标候选主体区域进行调整和分类,以获得至少一个目标候选主体区域集合;以及对每个目标候选主体区域集合中的目标候选主体区域合并,以识别出所述图像的主体区域。
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| US15/322,466 US10121095B2 (en) | 2014-12-30 | 2015-06-23 | Method and device for recognizing subject area of image |
| JP2017501172A JP6397986B2 (ja) | 2014-12-30 | 2015-06-23 | 画像オブジェクト領域の認識方法及び装置 |
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410850511.2A CN104573715B (zh) | 2014-12-30 | 2014-12-30 | 图像主体区域的识别方法及装置 |
| CN201410850511.2 | 2014-12-30 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| WO2016107103A1 true WO2016107103A1 (zh) | 2016-07-07 |
Family
ID=53089735
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| PCT/CN2015/082049 Ceased WO2016107103A1 (zh) | 2014-12-30 | 2015-06-23 | 图像主体区域的识别方法及装置 |
Country Status (4)
| Country | Link |
|---|---|
| US (1) | US10121095B2 (zh) |
| JP (1) | JP6397986B2 (zh) |
| CN (1) | CN104573715B (zh) |
| WO (1) | WO2016107103A1 (zh) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2018152532A1 (en) * | 2017-02-20 | 2018-08-23 | Alibaba Group Holding Linited | Type prediction method, apparatus and electronic device for recognizing an object in an image |
| JP2019530039A (ja) * | 2017-06-06 | 2019-10-17 | ミデア グループ カンパニー リミテッドMidea Group Co., Ltd. | ディープニューラルネットワークを用いた粗略から精細な手検出方法 |
| TWI764905B (zh) * | 2016-09-23 | 2022-05-21 | 南韓商三星電子股份有限公司 | 偵測物體的裝置與方法、製造處理器的方法以及建構積體電路的方法 |
| CN115797806A (zh) * | 2022-11-23 | 2023-03-14 | 华北电力大学(保定) | 一种输电线路多金具检测方法及装置 |
Families Citing this family (33)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104573715B (zh) * | 2014-12-30 | 2017-07-25 | 百度在线网络技术(北京)有限公司 | 图像主体区域的识别方法及装置 |
| CN104866565B (zh) * | 2015-05-21 | 2019-03-12 | 百度在线网络技术(北京)有限公司 | 图片搜索方法及装置 |
| CN104992146B (zh) * | 2015-06-15 | 2019-09-20 | 百度在线网络技术(北京)有限公司 | 一种用于人脸识别的方法和装置 |
| WO2017000118A1 (en) * | 2015-06-29 | 2017-01-05 | Xiaoou Tang | Method and apparatus for predicting attribute for image sample |
| CN106326817B (zh) * | 2015-07-03 | 2021-08-03 | 佳能株式会社 | 用于从图像中检测对象的方法和设备 |
| CN106803234B (zh) | 2015-11-26 | 2020-06-16 | 腾讯科技(深圳)有限公司 | 图片编辑中的图片显示控制方法及装置 |
| CN106203284B (zh) * | 2016-06-30 | 2019-05-14 | 华中科技大学 | 基于卷积神经网络和条件随机场的人脸检测方法 |
| CN107729910A (zh) * | 2016-08-11 | 2018-02-23 | 阿里巴巴集团控股有限公司 | 一种图形识别的方法、装置及系统 |
| CN107292293A (zh) * | 2017-07-26 | 2017-10-24 | 北京小米移动软件有限公司 | 人脸检测的方法及装置 |
| CN108062739B (zh) * | 2017-11-02 | 2021-07-02 | 广东数相智能科技有限公司 | 一种基于主体位置的图片智能裁剪方法及装置 |
| CN108229305B (zh) | 2017-11-21 | 2021-06-04 | 北京市商汤科技开发有限公司 | 用于确定目标对象的外接框的方法、装置和电子设备 |
| CN108132054A (zh) * | 2017-12-20 | 2018-06-08 | 百度在线网络技术(北京)有限公司 | 用于生成信息的方法和装置 |
| CN110569856B (zh) * | 2018-08-24 | 2020-07-21 | 阿里巴巴集团控股有限公司 | 样本标注方法及装置、损伤类别的识别方法及装置 |
| US11850514B2 (en) | 2018-09-07 | 2023-12-26 | Vulcan Inc. | Physical games enhanced by augmented reality |
| CN109492684A (zh) * | 2018-10-31 | 2019-03-19 | 西安同瑞恒达电子科技有限公司 | 数据处理方法及装置 |
| KR102701601B1 (ko) * | 2018-11-21 | 2024-09-03 | 삼성전자주식회사 | 영상 처리 장치 및 그 제어방법 |
| US11670080B2 (en) * | 2018-11-26 | 2023-06-06 | Vulcan, Inc. | Techniques for enhancing awareness of personnel |
| CN109801270B (zh) * | 2018-12-29 | 2021-07-16 | 北京市商汤科技开发有限公司 | 锚点确定方法及装置、电子设备和存储介质 |
| WO2020163530A1 (en) | 2019-02-08 | 2020-08-13 | Vulcan Inc. | Devices to assist ecosystem development and preservation |
| US11912382B2 (en) | 2019-03-22 | 2024-02-27 | Vulcan Inc. | Underwater positioning system |
| US11435845B2 (en) | 2019-04-23 | 2022-09-06 | Amazon Technologies, Inc. | Gesture recognition based on skeletal model vectors |
| WO2020237611A1 (zh) * | 2019-05-31 | 2020-12-03 | 深圳市大疆创新科技有限公司 | 图像处理方法、装置、控制终端及可移动设备 |
| JP7328012B2 (ja) * | 2019-06-03 | 2023-08-16 | Jx金属株式会社 | 電子・電気機器部品屑の組成解析方法、電子・電気機器部品屑の組成解析装置、及び電子・電気機器部品屑の処理方法 |
| JP7328011B2 (ja) * | 2019-06-03 | 2023-08-16 | Jx金属株式会社 | 電子・電気機器部品屑の組成解析方法、電子・電気機器部品屑の組成解析装置、及び電子・電気機器部品屑の処理方法 |
| CN110248097B (zh) * | 2019-06-28 | 2021-02-23 | Oppo广东移动通信有限公司 | 追焦方法、装置、终端设备、计算机可读存储介质 |
| CN110490196B (zh) * | 2019-08-09 | 2022-11-15 | Oppo广东移动通信有限公司 | 主体检测方法和装置、电子设备、计算机可读存储介质 |
| CN110580678B (zh) * | 2019-09-10 | 2023-06-20 | 北京百度网讯科技有限公司 | 图像处理方法及装置 |
| CN110796041B (zh) | 2019-10-16 | 2023-08-18 | Oppo广东移动通信有限公司 | 主体识别方法和装置、电子设备、计算机可读存储介质 |
| CN111340124A (zh) * | 2020-03-03 | 2020-06-26 | Oppo广东移动通信有限公司 | 图像中实体类别的识别方法和装置 |
| CN113297514B (zh) * | 2020-04-13 | 2025-08-12 | 阿里巴巴集团控股有限公司 | 图像处理方法、装置、电子设备和计算机存储介质 |
| CN112232215B (zh) * | 2020-10-16 | 2021-04-06 | 哈尔滨市科佳通用机电股份有限公司 | 一种铁路货车钩尾销托梁脱落故障检测方法 |
| CN113222041B (zh) * | 2021-05-24 | 2022-06-07 | 北京航空航天大学 | 图结构表征的高阶关联发现细粒度图像识别方法及装置 |
| KR102720959B1 (ko) * | 2021-06-01 | 2024-10-24 | (주)애니로직 | 웹 디자인 속성을 추출하는 웹 문서 생성 장치 및 방법 |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101894361A (zh) * | 2010-07-01 | 2010-11-24 | 云南大学 | 一种基于内容的主体大小能控的图像缩放方法 |
| CN102013021A (zh) * | 2010-08-19 | 2011-04-13 | 汪建 | 基于颜色和区域生长的茶叶嫩芽分割和识别方法 |
| CN102622761A (zh) * | 2012-04-13 | 2012-08-01 | 西安电子科技大学 | 基于相似性相互作用机理的图像分割方法 |
| CN104573715A (zh) * | 2014-12-30 | 2015-04-29 | 百度在线网络技术(北京)有限公司 | 图像主体区域的识别方法及装置 |
Family Cites Families (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2004296921A (ja) * | 2003-03-27 | 2004-10-21 | Canon Inc | 位置検出装置 |
| CN100468453C (zh) * | 2003-12-29 | 2009-03-11 | 佳能株式会社 | 检测和校正图像中的红眼的方法 |
| US7899644B2 (en) * | 2004-02-05 | 2011-03-01 | Bae Systems Information And Electronic Systems Integration Inc. | Threat launch detection system and method |
| WO2006030633A1 (ja) * | 2004-09-15 | 2006-03-23 | Matsushita Electric Industrial Co., Ltd. | 物体検出装置 |
| WO2008065717A1 (fr) * | 2006-11-29 | 2008-06-05 | Fujitsu Limited | Système et procédé de détection de piéton |
| KR101030430B1 (ko) * | 2007-09-12 | 2011-04-20 | 주식회사 코아로직 | 영상 처리 장치와 방법 및 그 기록매체 |
| CN101408942B (zh) * | 2008-04-17 | 2011-01-12 | 浙江师范大学 | 一种复杂背景下的车牌定位方法 |
| JP2010039968A (ja) * | 2008-08-08 | 2010-02-18 | Hitachi Ltd | オブジェクト検出装置及び検出方法 |
| CN101350063B (zh) * | 2008-09-03 | 2011-12-28 | 北京中星微电子有限公司 | 人脸特征点定位方法及装置 |
| CN101510309B (zh) * | 2009-03-30 | 2010-12-01 | 西安电子科技大学 | 基于复小波纹理区域合并的改进分水岭sar图像分割方法 |
| JP5702960B2 (ja) * | 2010-07-12 | 2015-04-15 | キヤノン株式会社 | 画像処理装置、画像処理方法、及びプログラム |
| JP2013011950A (ja) * | 2011-06-28 | 2013-01-17 | Canon Inc | 画像処理装置、画像処理方法及びプログラム |
| US9916538B2 (en) * | 2012-09-15 | 2018-03-13 | Z Advanced Computing, Inc. | Method and system for feature detection |
| CN103955718A (zh) * | 2014-05-15 | 2014-07-30 | 厦门美图之家科技有限公司 | 一种图像主体对象的识别方法 |
-
2014
- 2014-12-30 CN CN201410850511.2A patent/CN104573715B/zh active Active
-
2015
- 2015-06-23 WO PCT/CN2015/082049 patent/WO2016107103A1/zh not_active Ceased
- 2015-06-23 JP JP2017501172A patent/JP6397986B2/ja not_active Expired - Fee Related
- 2015-06-23 US US15/322,466 patent/US10121095B2/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101894361A (zh) * | 2010-07-01 | 2010-11-24 | 云南大学 | 一种基于内容的主体大小能控的图像缩放方法 |
| CN102013021A (zh) * | 2010-08-19 | 2011-04-13 | 汪建 | 基于颜色和区域生长的茶叶嫩芽分割和识别方法 |
| CN102622761A (zh) * | 2012-04-13 | 2012-08-01 | 西安电子科技大学 | 基于相似性相互作用机理的图像分割方法 |
| CN104573715A (zh) * | 2014-12-30 | 2015-04-29 | 百度在线网络技术(北京)有限公司 | 图像主体区域的识别方法及装置 |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| TWI764905B (zh) * | 2016-09-23 | 2022-05-21 | 南韓商三星電子股份有限公司 | 偵測物體的裝置與方法、製造處理器的方法以及建構積體電路的方法 |
| WO2018152532A1 (en) * | 2017-02-20 | 2018-08-23 | Alibaba Group Holding Linited | Type prediction method, apparatus and electronic device for recognizing an object in an image |
| US10706334B2 (en) | 2017-02-20 | 2020-07-07 | Alibaba Group Holding Limited | Type prediction method, apparatus and electronic device for recognizing an object in an image |
| TWI746674B (zh) * | 2017-02-20 | 2021-11-21 | 香港商阿里巴巴集團服務有限公司 | 識別圖像中物件的類型預測方法、裝置及電子設備 |
| JP2019530039A (ja) * | 2017-06-06 | 2019-10-17 | ミデア グループ カンパニー リミテッドMidea Group Co., Ltd. | ディープニューラルネットワークを用いた粗略から精細な手検出方法 |
| US10817716B2 (en) | 2017-06-06 | 2020-10-27 | Midea Group Co., Ltd. | Coarse-to-fine hand detection method using deep neural network |
| CN115797806A (zh) * | 2022-11-23 | 2023-03-14 | 华北电力大学(保定) | 一种输电线路多金具检测方法及装置 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104573715A (zh) | 2015-04-29 |
| US20170293824A1 (en) | 2017-10-12 |
| US10121095B2 (en) | 2018-11-06 |
| JP6397986B2 (ja) | 2018-09-26 |
| CN104573715B (zh) | 2017-07-25 |
| JP2017520859A (ja) | 2017-07-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| JP6397986B2 (ja) | 画像オブジェクト領域の認識方法及び装置 | |
| Kakani et al. | Improved OCR based automatic vehicle number plate recognition using features trained neural network | |
| CN106570453B (zh) | 用于行人检测的方法、装置和系统 | |
| US9501703B2 (en) | Apparatus and method for recognizing traffic sign board | |
| Qin et al. | Cascaded segmentation-detection networks for word-level text spotting | |
| KR102195940B1 (ko) | 적응적 비최대억제 방법을 이용하는 딥러닝기반 영상객체 탐지를 위한 장치 및 방법 | |
| WO2019169532A1 (zh) | 车牌识别方法及云系统 | |
| CN101872477A (zh) | 检测图像中的对象的方法、装置,及包括该装置的系统 | |
| CN105512683A (zh) | 基于卷积神经网络的目标定位方法及装置 | |
| Kulkarni et al. | Text detection and recognition: a review | |
| Setiyono et al. | Number plate recognition on vehicle using YOLO-Darknet | |
| CN104318208A (zh) | 一种基于图分割和实例学习的视频场景检测方法 | |
| Uddin et al. | Horse detection using haar like features | |
| Roslan et al. | Color-based bird image classification using Support Vector Machine | |
| CN104537392B (zh) | 一种基于判别性语义部件学习的对象检测方法 | |
| Pham et al. | CNN-based character recognition for license plate recognition system | |
| Reddy et al. | Real Time Object Identification: A Study on COCO Dataset | |
| Xue | Optical character recognition | |
| KR102050422B1 (ko) | 문자 인식 장치 및 방법 | |
| CN116433982B (zh) | 一种基于自适应域学习的弱监督目标定位方法 | |
| CN110121723A (zh) | 人工神经网络 | |
| Muchtar et al. | Attention-based approach for efficient moving vehicle classification | |
| CN104680123A (zh) | 对象识别装置、对象识别方法和程序 | |
| Göring et al. | Semantic Segmentation using GrabCut. | |
| Gao et al. | A hierarchical visual saliency model for character detection in natural scenes |
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: 15874793 Country of ref document: EP Kind code of ref document: A1 |
|
| WWE | Wipo information: entry into national phase |
Ref document number: 15322466 Country of ref document: US |
|
| ENP | Entry into the national phase |
Ref document number: 2017501172 Country of ref document: JP Kind code of ref document: A |
|
| NENP | Non-entry into the national phase |
Ref country code: DE |
|
| 122 | Ep: pct application non-entry in european phase |
Ref document number: 15874793 Country of ref document: EP Kind code of ref document: A1 |