WO2015096565A1 - 图像中的目标物的识别方法及装置 - Google Patents

图像中的目标物的识别方法及装置 Download PDF

Info

Publication number
WO2015096565A1
WO2015096565A1 PCT/CN2014/090976 CN2014090976W WO2015096565A1 WO 2015096565 A1 WO2015096565 A1 WO 2015096565A1 CN 2014090976 W CN2014090976 W CN 2014090976W WO 2015096565 A1 WO2015096565 A1 WO 2015096565A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
unknown
classification
classifier
category
Prior art date
Application number
PCT/CN2014/090976
Other languages
English (en)
French (fr)
Inventor
江焯林
宋阳秋
杨强
Original Assignee
华为技术有限公司
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 华为技术有限公司 filed Critical 华为技术有限公司
Publication of WO2015096565A1 publication Critical patent/WO2015096565A1/zh
Priority to US15/193,209 priority Critical patent/US9798956B2/en

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/35—Categorising the entire scene, e.g. birthday party or wedding scene
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/217—Validation; Performance evaluation; Active pattern learning techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR 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
    • G06V10/762—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using clustering, e.g. of similar faces in social networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR 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
    • G06V10/77—Processing 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
    • G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR 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
    • G06V10/77—Processing 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
    • G06V10/776—Validation; Performance evaluation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/70—Labelling scene content, e.g. deriving syntactic or semantic representations

Definitions

  • the present invention relates to the field of computer technologies, and in particular, to a method and an apparatus for identifying a target in an image.
  • Image object detection and recognition is used to identify people or objects in the image, and classify the image (for example, mark the areas in the image as "sky”, “beach”, “sun”, etc.), one of which is the most
  • a typical problem is to identify a certain type of object in an image, such as the Caltech 101 data set, which is a similar problem.
  • Image target detection and recognition is one of the core issues in the field of computer vision, and one of the important breakthroughs in the field of artificial intelligence.
  • the current target detection methods are mostly by fixing a certain type of object, by modeling its shape or edge (or even the bounding box), by scanning the position of the object in the image and fitting it.
  • Edge detection can be obtained using methods such as the Canny operator; shape or edge modeling and tracking can be obtained using methods such as Condensation, Kalman filter, or Meanshift.
  • the target detection mostly judges a certain type of object (such as a human face, a human body, a certain type of specific object, etc.), the understanding of the unknown object has not been involved. If a new target does not appear in need Inside the object, it is difficult to judge.
  • a certain type of object such as a human face, a human body, a certain type of specific object, etc.
  • Embodiments of the present invention provide a method and a device for identifying an object in an image, which can identify an object of an unknown classification.
  • a first aspect of the present invention provides a method for identifying a target in an image, which may include:
  • Knowing classifications included in the automatic clustering result are classified into corresponding known classifications to identify known objects in the image;
  • the classifier is trained in a machine learning manner to identify an unknown target in the image.
  • the unknown classifier included in the automatic clustering result is trained by a machine learning manner to perform an unknown target object in the image.
  • Identification can include:
  • the classifier is trained on the unknown classification with the category annotation to identify the unknown target in the image.
  • the performing category labeling includes:
  • the classification is trained by the migration learning to classify the unknown classification with the category label to identify the unknown in the image.
  • Target identification including:
  • the migration learning method and the updated existing classifier are used to train the classifier for the unknown classification with the category annotation to identify unknown objects in the image.
  • the method further includes:
  • a second aspect of the present invention provides an image processing apparatus, which may include:
  • a feature acquisition module configured to extract feature data from the image, and convert the extracted feature data into a unified expression
  • An automatic clustering module configured to automatically cluster features in the image according to the feature data and the historical clustering result uniformly expressed by the feature acquiring module;
  • a first classification module configured to classify, as a known classification included in the automatic clustering result, a corresponding known classification to identify a known target in the image
  • a second classification module configured to train the classifier in a machine learning manner to identify an unknown target in the image for an unknown classification included in the automatic clustering result.
  • the second classification module includes:
  • An annotation module configured to perform category labeling on an unknown classification included in the automatic clustering result
  • a classification learning module for learning the unknown classification of the category labeling by migration learning
  • a classifier is trained to identify unknown objects in the image.
  • the labeling module is specifically configured to obtain the category labeling information input by the user by means of human-computer interaction; or, searching and searching from the Internet
  • the unknown classification similarity reaches the specified requirement image, and the unknown classification is marked by the annotation information of the image through the Internet.
  • the classification learning module is specifically configured to update an existing classifier according to the result of the automatic clustering, and use the migration
  • the learning method and the updated existing classifier train the classifier for the unknown classification with the category annotation to identify unknown objects in the image.
  • the automatic clustering module is further configured to update the location according to the automatic clustering result.
  • the historical clustering results are further configured to update the location according to the automatic clustering result.
  • feature data is extracted from an image, and the extracted feature data is subjected to expression processing; according to the feature data and the historical clustering result after the expression processing,
  • the features in the image are automatically clustered;
  • the known classifications included in the automatic clustering results are classified into corresponding known classifications to identify known objects in the image;
  • An unknown classification included in the automatic clustering result is trained by a machine learning method to identify an unknown target in the image. Therefore, in the embodiment of the present invention, when the target object that does not belong to the existing category is included in the image to be identified, the target object that does not belong to the existing category is learned, and a new classifier is obtained, thereby achieving the target of the unknown classification. The object is identified.
  • FIG. 1 is a schematic flow chart of an embodiment of a method for identifying an object in an image according to the present invention
  • FIG. 2 is a schematic flow chart of an embodiment of step S104 in FIG. 1;
  • FIG. 3 is a schematic structural diagram of an embodiment of an image processing apparatus according to the present invention.
  • FIG. 4 is a schematic structural diagram of an embodiment of a second classification module in FIG. 3 according to the present invention.
  • Fig. 5 is a block diagram showing the structure of another embodiment of the image processing apparatus of the present invention.
  • FIG. 1 is a flow chart showing an embodiment of a method for identifying a target in an image of the present invention. As shown in Figure 1, it can include the following steps:
  • the feature data of the image according to the embodiment of the present invention includes, but is not limited to, a geometric feature, a shape feature, a color feature, a texture feature, and the like.
  • a Canny operator, a Laplacian or a Laplacian of Gassian (LOG) operator may be used to extract edge features of the image; and Singular Value Decomposition (Singular Value Decomposition, SVD) algorithm extracts the texture features of the image; using the Histogram of Oriented Gradient (HOG) descriptor or Scale-Invariant Feature Transform (SIFT) algorithm to obtain the feature vector of the image; using principal component analysis (Principal Component Analysis, PCA) algorithm, Linear Discriminant Analysis (LDA) algorithm or Independent components analysis (ICA) algorithm to extract global or local features of images.
  • PCA Principal Component Analysis
  • LDA Linear Discriminant Analysis
  • ICA Independent components analysis
  • PCA reconstructs samples by using a low-dimensional feature vector and projection matrix to model feature vectors by minimizing reconstruction errors. At the same time, it leaves the dimension with large variance and the dimension with small variance removed. Removing the small variance variance can help the sample space reduce the uncertainty, leaving a dimension with a large variance to maintain the local distance between the sample and the sample.
  • S102 Perform automatic clustering on features in the image according to the uniformly expressed feature data and historical clustering results.
  • the automatic clustering of the present invention may refer to an unsupervised classification, that is, without any prior knowledge, without knowing the category of each image in each image to be recognized in advance, and according to the characteristics of each image to be recognized.
  • Classification which divides images with similar or identical features into the same subclass.
  • the number of subclasses cannot be known in advance, and a clustering analysis based on a probability distribution model may be used, such as a Dirichlet Processes Clustering algorithm; or a Canopy clustering is first used.
  • the algorithm performs preprocessing, and then uses the partition-based method for cluster analysis, such as K-Means clustering algorithm.
  • the historical clustering result may also be updated according to an automatic clustering result.
  • Step S103 classifying the known classifications included in the automatic clustering result into corresponding known classifications to identify known objects in the image.
  • Step S104 For the unknown classification included in the automatic clustering result, the classifier is trained by means of machine learning to identify the unknown target in the image.
  • step S104 may further include:
  • Step S1041 performing category labeling on the unknown classification included in the automatic clustering result.
  • Step S1042 Train the classifier for the unknown classification with the category label by migration learning to identify the unknown target in the image.
  • step S1041 the performing category labeling includes:
  • an image matching the unknown classification similarity to the specified requirement is searched from the Internet, and the unknown classification is classified by the annotation information of the image through the Internet.
  • the result of the automatic clustering may be marked by the user's category of "alpine".
  • the unknown classification can be classified by "mountain” on the Internet. Label.
  • step S1042 by using the migration learning, the unknown classification training classifier with the category labeling is used to identify the unknown target object in the image, and specifically according to the result of the automatic clustering. Updating the existing classifier; using the migration learning method and the updated existing classifier to train the classifier for the unknown classification with the category label to perform unknown target in the image Identification.
  • the existing and trained classifiers may be a Support Vector Machine (SVM) classifier, a Bayesian classifier, a decision tree classifier, and a naive Bayes classifier (Naive).
  • SVM Support Vector Machine
  • Bayesian classifier Bayesian classifier
  • decision tree classifier decision tree classifier
  • Naive naive Bayes classifier
  • Bayes Classifier NBC
  • the migration learning method in the embodiment of the present invention includes, but is not limited to, a covariance shift, a TrAdaboost, a multi-task based learning, and the like.
  • the migration learning mode when adopted, as the time increases, the data increases, the learning starting point is higher, the convergence speed is faster, and the trained classifier is better. Moreover, it can update the historical clustering with new clustering results, and update the existing classifier through the automatic clustering result, thereby realizing continuous updating and continuous learning of the entire system, thereby continuously optimizing the system.
  • feature data is extracted from an image, And performing expression processing on the extracted feature data; automatically clustering features in the image according to the feature data and the historical clustering result after the expression processing; and the known features included in the automatic clustering result Classification, classified into corresponding known classifications to identify known objects in the image; for unknown classifications included in the automatic clustering results, the classifier is trained by machine learning to An unknown target in the image is identified. Therefore, in the embodiment of the present invention, when the target object that does not belong to the existing category is included in the image to be identified, the target object that does not belong to the existing category is learned, and a new classifier is obtained, thereby achieving the target of the unknown classification. The object is identified.
  • an embodiment of the present invention also provides an image processing apparatus that can be used to implement a method of identifying an object in an image of the present invention.
  • FIG. 3 is a schematic structural view showing an embodiment of an image processing apparatus according to the present invention. As shown in FIG. 3, it may include: a feature acquisition module 31, an automatic clustering module 32, a first classification module 33, and a second classification module 34, wherein:
  • a feature acquiring module 31 configured to extract feature data from an image, and convert the extracted feature data into a unified expression
  • the automatic clustering module 32 is configured to automatically cluster the features in the image according to the feature data and the historical clustering result uniformly expressed by the feature acquiring module 31;
  • the first classification module 33 is configured to classify the known classification included in the result of the automatic clustering by the automatic clustering module 32 into a corresponding known classification to perform the known target in the image. Identification
  • the second classification module 34 is configured to train the classifier in a machine learning manner for the unknown classification included in the result of the automatic clustering module 32 to automatically identify the unknown target in the image.
  • the feature data of the image in the embodiment of the present invention includes, but is not limited to, a geometric feature, a shape feature, a color feature, a texture feature, and the like.
  • the feature obtaining module 31 may use a Canny operator, a Laplacian or a Laplacian of Gassian (LOG) operator to extract edge features of the image; and use Singular Value Decomposition (Singular Value Decomposition, SVD) algorithm extracts the texture features of the image; using the Histogram of Oriented Gradient (HOG) descriptor or Scale-Invariant Feature Transform (SIFT) algorithm to obtain the feature vector of the image; using principal component analysis (Principal Component Analysis, PCA) algorithm, Linear Discriminant Analysis (LDA) algorithm or Independent components analysis (ICA) algorithm to extract global or local features of images. In order to achieve denoising and improve the recognition effect.
  • PCA Principal Component Analysis
  • LDA Linear Discriminant Analysis
  • ICA Independent components analysis
  • PCA reconstructs samples by using a low-dimensional feature vector and projection matrix to model feature vectors by minimizing reconstruction errors. At the same time, it leaves the dimension with large variance and the dimension with small variance removed. Removing the small variance variance can help the sample space reduce the uncertainty, leaving a dimension with a large variance to maintain the local distance between the sample and the sample.
  • the automatic clustering module 32 may not need any prior knowledge, and does not know the category of each image in each image to be recognized in advance, and classifies according to the characteristics of each image to be identified, and will have similar or Images of the same feature are divided into the same subclass.
  • the automatic clustering module 32 may perform cluster analysis based on a probability distribution model, such as a Dirichlet Processes Clustering algorithm; or adopt Canopy first.
  • the clustering algorithm performs preprocessing, and then uses the partition-based method for cluster analysis, such as K-Means clustering algorithm.
  • the automatic clustering module 32 may further update the historical clustering result according to the automatic clustering result.
  • the second classification module 34 may further include:
  • An annotation module 341, configured to perform a category on an unknown classification included in the automatic clustering result Label
  • the classification learning module 342 is configured to train the classifier for the unknown classification with the category annotation by migration learning to identify the unknown target in the image.
  • the labeling module 341 is specifically configured to obtain the category labeling information input by the user by means of human-computer interaction; or, by using the Internet, to search for an image that meets the specified requirement with the unknown classification similarity, and An annotation information of the image, and classifying the unknown classification.
  • the result of the automatic clustering may be marked by the user's category of "alpine".
  • the unknown classification can be classified by "mountain” on the Internet. Label.
  • the classification learning module 342 is specifically configured to update an existing classifier according to the result of the automatic clustering, and use the migration learning method and the updated existing classifier as the An unknown classification of the category annotation trains the classifier to identify unknown objects in the image.
  • the existing and trained classifiers may be a Support Vector Machine (SVM) classifier, a Bayesian classifier, a decision tree classifier, and a naive Bayes classifier (Naive). Bayes Classifier, NBC), etc.
  • the migration learning method in the embodiment of the present invention includes, but is not limited to, a covariance shift, a TrAdaboost, a multi-task based learning, and the like.
  • the migration learning mode when adopted, as the time increases, the data increases, the learning starting point is higher, the convergence speed is faster, and the trained classifier is better. Moreover, it can update the historical clustering with new clustering results, and update the existing classifier through the automatic clustering result, thereby realizing continuous updating and continuous learning of the entire system, thereby continuously optimizing the system.
  • feature data is extracted from an image, and the extracted feature data is subjected to expression processing; according to the feature data and the historical clustering result after the expression processing,
  • the features in the image are automatically clustered;
  • the known classifications included in the automatic clustering results are classified into corresponding known classifications to identify known objects in the image;
  • An unknown classification included in the automatic clustering result is trained by a machine learning method to identify an unknown target in the image. Therefore, in the embodiment of the present invention, when the target object that does not belong to the existing category is included in the image to be identified, the target object that does not belong to the existing category is learned, and a new classifier is obtained, thereby achieving the target of the unknown classification. The object is identified.
  • Fig. 5 is a block diagram showing the structure of another embodiment of the image processing apparatus of the present invention. As shown in FIG. 5, it may include a memory 51 and a processor 52, wherein the processor 52 calls the program code stored in the memory 51 and performs the following steps:
  • Knowing classifications included in the automatic clustering result are classified into corresponding known classifications to identify known objects in the image;
  • the classifier is trained in a machine learning manner to identify an unknown target in the image.
  • the processor performs a step of training the classifier by machine learning to identify an unknown target in the image for an unknown classification included in the automatic clustering result , can perform the following steps specifically:
  • the processor 52 performs category labeling, including:
  • an image matching the unknown classification similarity to the specified requirement is searched from the Internet, and the unknown classification is classified by the annotation information of the image through the Internet.
  • the processor 52 through the migration learning, trains the classifier for the unknown classification with the category label to identify the unknown target in the image, and performs the following steps. :
  • the migration learning method and the updated existing classifier are used to train the classifier for the unknown classification with the category annotation to identify unknown objects in the image.
  • the method further includes:
  • the modules of the embodiments of the present invention may be implemented by a general-purpose integrated circuit (such as a central processing unit CPU) or by an application specific integrated circuit (ASIC).
  • a person skilled in the art can understand that all or part of the steps of the foregoing embodiments can be completed by a program, and the program can be stored in a computer readable storage medium, and the storage medium can include: A disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

Landscapes

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

Abstract

一种图像中的目标物的识别方法及装置，其中所述方法包括：从图像中提取特征数据，并将所述提取的特征数据转换为统一的表达；根据所述统一表达后的特征数据和历史聚类结果，对图像中的特征进行自动聚类；对于所述自动聚类结果中包括的已知的分类，归类为对应的已知分类，以对所述图像中的已知目标物进行识别；对于所述自动聚类结果中包括的未知分类，通过机器学习的方式训练分类器，以对所述图像中的未知目标物进行识别。可实现对未知分类的目标物进行识别。

Description

图像中的目标物的识别方法及装置
本申请要求于2013年12月27日提交中国专利局，申请号为201310739555.3、发明名称为“图像中的目标物的识别方法及装置”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本发明涉及计算机技术领域，尤其涉及一种图像中的目标物的识别方法及装置。
背景技术
数字图像理解包含几个层次，如图像分割、边缘检测、图像目标检测和识别等。其中图像目标检测和识别是用来识别图像中的人或物体，对图像进行类别标注(如，将图像中区域分别标注为“天空”、“海滩”、“太阳”等)，其中一类最典型的问题是识别图像中某一类型的物体，如Caltech101数据集即为类似的问题。图像目标检测和识别是计算机视觉领域的核心问题之一，也是人工智能领域的重要突破口之一。
目前的目标检测方法多是通过固定某一类物体，通过对其形状或边缘(甚至bounding box)进行建模，通过扫描图像中物体的位置并进行拟合得到。边缘检测可以使用Canny算子等方法获得；形状或边缘建模和跟踪可以使用Condensation、Kalman filter或Meanshift等方法获得。
由于目标检测多对于已知一类物体(如人脸、人体、某类特定物体等)进行判断，对未知物体的理解还没有涉及。如果新来一个目标没有出现在需要跟 踪的对象里面，则很难进行判断。
发明内容
本发明实施例提供一种图像中的目标物的识别方法及装置，可对未知分类的目标物进行识别。
本发明第一方面提供一种图像中的目标物的识别方法，其可包括：
从图像中提取特征数据，并将所述提取的特征数据转换为统一的表达；
根据所述统一表达后的特征数据和历史聚类结果，对图像中的特征进行自动聚类；
对于所述自动聚类结果中包括的已知的分类，归类为对应的已知分类，以对所述图像中的已知目标物进行识别；
对于所述自动聚类结果中包括的未知分类，通过机器学习的方式训练分类器，以对所述图像中的未知目标物进行识别。
结合第一方面，在第一种可行的实施方式中，所述对于所述自动聚类结果中包括的未知分类，通过机器学习的方式训练分类器，以对所述图像中的未知目标物进行识别，可包括：
对于所述自动聚类结果中包括的未知的分类，进行类别标注；
通过迁移学习，对所述带有类别标注的未知的分类训练分类器，以对所述图像中的未知目标物进行识别。
结合第一方面的第一种可行的实施方式，在第二种可行的实施方式中，所述进行类别标注，包括：
通过人机交互的方式，获取用户输入的类别标注信息；
或者，从互联网查找与所述未知的分类相似性达到指定要求的图像，并通 过互联网对所述图像的标注信息，对所述未知的分类进行类别标注。
结合第一方面的第一种可行的实施方式，在第三种可行的实施方式中，通过迁移学习，对所述带有类别标注的未知的分类训练分类器，以对所述图像中的未知目标物进行识别，包括：
根据自动聚类的结果，对已有的分类器进行更新；
使用迁移学习方法和所述更新后的已有的分类器为所述带有类别标注的未知的分类训练分类器，以对所述图像中的未知目标物进行识别。
结合第一方面至第一方面的第三种可行的实施方式中任一种，在第四种可行的实施方式中，所述对图像中的特征进行自动聚类之后，还包括：
根据所述自动聚类结果，更新所述历史聚类结果。
本发明第二方面提供一种图像处理装置，其可包括：
特征获取模块，用于从图像中提取特征数据，并将所述提取的特征数据转换为统一的表达；
自动聚类模块，用于根据所述特征获取模块统一表达后的特征数据和历史聚类结果，对图像中的特征进行自动聚类；
第一分类模块，用于对于所述自动聚类结果中包括的已知的分类，归类为对应的已知分类，以对所述图像中的已知目标物进行识别；
第二分类模块，用于对于所述自动聚类结果中包括的未知分类，通过机器学习的方式训练分类器，以对所述图像中的未知目标物进行识别。
结合第二方面，在第一种可行的实施方式中，所述第二分类模块，包括：
标注模块，用于对于所述自动聚类结果中包括的未知的分类，进行类别标注；
分类学习模块，用于通过迁移学习，对所述带有类别标注的未知的分类训 练分类器，以对所述图像中的未知目标物进行识别。
结合第二方面的第一种可行的实施方式，在第二种可行的实施方式中，标注模块具体用于通过人机交互的方式，获取用户输入的类别标注信息；或者，从互联网查找与所述未知的分类相似性达到指定要求的图像，并通过互联网对所述图像的标注信息，对所述未知的分类进行类别标注。
结合第二方面的第一种可行的实施方式，在第三种可行的实施方式中，所述分类学习模块具体用于根据自动聚类的结果，对已有的分类器进行更新，并使用迁移学习方法和所述更新后的已有的分类器为所述带有类别标注的未知的分类训练分类器，以对所述图像中的未知目标物进行识别。
结合第二方面至第二方面的第三种可行的实施方式中任一种，在第四种可行的实施方式中，所述自动聚类模块还用于根据所述自动聚类结果，更新所述历史聚类结果。
由上可见，在本发明的一些可行的实施方式中，从图像中提取特征数据，并对所述提取的特征数据进行表达处理；根据所述表达处理后的特征数据和历史聚类结果，对图像中的特征进行自动聚类；对于所述自动聚类结果中包括的已知的分类，归类为对应的已知分类，以对所述图像中的已知目标物进行识别；对于所述自动聚类结果中包括的未知分类，通过机器学习的方式训练分类器，以对所述图像中的未知目标物进行识别。由此，本发明实施例可在待识别的图像中包含不属于已有类别的目标物时，对不属于已有类别的目标物进行学习，得到新的分类器，从而实现对未知分类的目标物进行识别。
附图说明
图1为本发明的图像中的目标物的识别方法的一实施例的流程示意图；
图2为图1中步骤S104的一实施例的流程示意图；
图3为本发明的图像处理装置的一实施例的结构组成示意图；
图4为本发明图3中的第二分类模块的一实施例的结构组成示意图；
图5为本发明的图像处理装置的另一实施例的结构组成示意图。
具体实施方式
为使本发明的目的、技术方案和优点更加清楚，下面将结合附图对本发明作进一步的详细描述。
图1为本发明的图像中的目标物的识别方法的一实施例的流程示意图。如图1所示，其可包括以下步骤：
S101，从图像中提取特征数据，并将所述提取的特征数据转换为统一的表达。
在一些可行的实施方式中，本发明实施例所述的图像的特征数据包括但不限于：几何特征、形状特征、颜色特征、纹理特征等。
具体实现中，步骤S101中，可采用Canny算子、拉普拉斯算子或拉普拉斯高斯(Laplacian of Gassian，LOG)算子提取图像的边缘特征；采用奇异值分解(Singular Value Decomposition，SVD)算法提取图像的纹理特征；采用方向梯度直方图(Histogram of Oriented Gradient，HOG)描述子或尺度不变特征转换(Scale-Invariant Feature Transform，SIFT)算法得到图像的特征向量；采用主成分分析(Principal Component Analysis，PCA)算法、线性判别分析(Linear Discriminant Analysis，LDA)算法或独立成分分析(Independent components analysis，ICA)算法等提取图像的全局或局部特征等。以达到去噪、提高识别效果的作用。例如， PCA是通过使用一个低维的特征向量来和投影矩阵来重建样本，通过最小化重建误差来对特征向量进行建模。同时，它是把方差大的维度留下，方差小的维度去掉。去掉方差小的维度可以帮助样本空间减小不确定性，留下方差大的维度可以保持样本和样本之间的局部距离。
S102，根据所述统一表达后的特征数据和历史聚类结果，对图像中的特征进行自动聚类。
具体实现中，本发明的自动聚类可指无监督分类，即不需要任何先验知识，事先不了解各待识别的图像中的每一个图像的类别，而根据各待识别的图像的特征进行分类，将具有相似或相同特征的图像划分到同一子类。
在一些可行的实施方式中，无法预先得知子类的个数，可采用基于概率分布模型的方法进行聚类分析，如狄利克雷过程聚类(Dirichlet Processes Clustering)算法；或先采用Canopy聚类算法进行预处理，再采用基于划分的方法进行聚类分析，如K-均值(K-Means)聚类算法。
在一些可行的实施方式中，还可根据自动聚类结果，更新所述历史聚类结果。
步骤S103，对于所述自动聚类结果中包括的已知的分类，归类为对应的已知分类，以对所述图像中的已知目标物进行识别。
步骤S104，对于所述自动聚类结果中包括的未知分类，通过机器学习的方式训练分类器，以对所述图像中的未知目标物进行识别。
在一些可行的实施方式中，如图2，步骤S104可进一步包括：
步骤S1041，对于所述自动聚类结果中包括的未知的分类，进行类别标注。步骤S1042，通过迁移学习，对所述带有类别标注的未知的分类训练分类器，以对所述图像中的未知目标物进行识别。
具体实现中，在步骤S1041，所述进行类别标注，包括：
通过人机交互的方式，获取用户输入的类别标注信息；
或者，从互联网查找与所述未知的分类相似性达到指定要求的图像，并通过互联网对所述图像的标注信息，对所述未知的分类进行类别标注。
比如，假设已知类别中，不包括“高山”这样的类别，则在步骤S104，对于自动聚类的结果，可通过用户给予“高山”的类别标注。或者，通过查找互联网，发现局聚类结果中的某一未知的分类的相似度很高的物体被称之为“高山”，则可采用互联网上的“高山”对所述未知的分类进行类别标注。
具体实现中，在步骤S1042，通过迁移学习，对所述带有类别标注的未知的分类训练分类器，以对所述图像中的未知目标物进行识别时，具体可根据自动聚类的结果，对已有的分类器进行更新；使用迁移学习方法和所述更新后的已有的分类器为所述带有类别标注的未知的分类训练分类器，以对所述图像中的未知目标物进行识别。
在一些可行的实施方式中，已有和训练出的分类器可以是支持向量机(Support Vector Machine，SVM)分类器、贝叶斯分类器、决策树分类器、朴素贝叶斯分类器(Naive Bayes Classifier，NBC)等。
具体实现中，本发明实施例所述的迁移学习方法包括但不限于：covariance shift，TrAdaboost，基于多任务的学习等方法。
本发明实施例，当采用迁移学习方式后，其随着时间的增加，数据的增加，使学习的起点更高，收敛速度更快，训练出的分类器更优。并且，其可用新的聚类结果更新历史聚类，以及通过自动聚类结果更新已有的分类器，由此可实现整个系统的不断更新和不断学习，进而使系统不断优化。
由上可见，在本发明的一些可行的实施方式中，从图像中提取特征数据， 并对所述提取的特征数据进行表达处理；根据所述表达处理后的特征数据和历史聚类结果，对图像中的特征进行自动聚类；对于所述自动聚类结果中包括的已知的分类，归类为对应的已知分类，以对所述图像中的已知目标物进行识别；对于所述自动聚类结果中包括的未知分类，通过机器学习的方式训练分类器，以对所述图像中的未知目标物进行识别。由此，本发明实施例可在待识别的图像中包含不属于已有类别的目标物时，对不属于已有类别的目标物进行学习，得到新的分类器，从而实现对未知分类的目标物进行识别。
相应的，本发明实施例还提供了一种可用于实施本发明的图像中的目标物的识别方法的图像处理装置。
图3为本发明的图像处理装置的一实施例的结构组成示意图。如图3所示，其可包括：特征获取模块31、自动聚类模块32、第一分类模块33以及第二分类模块34，其中：
特征获取模块31，用于从图像中提取特征数据，并将所述提取的特征数据转换为统一的表达；
自动聚类模块32，用于根据所述特征获取模块31统一表达后的特征数据和历史聚类结果，对图像中的特征进行自动聚类；
第一分类模块33，用于对于所述自动聚类模块32自动聚类的结果中包括的已知的分类，归类为对应的已知分类，以对所述图像中的已知目标物进行识别；
第二分类模块34，用于对于所述自动聚类模块32自动聚类的结果中包括的未知分类，通过机器学习的方式训练分类器，以对所述图像中的未知目标物进行识别。
具体实现中，本发明实施例所述的图像的特征数据包括但不限于：几何特征、形状特征、颜色特征、纹理特征等。
具体实现中，特征获取模块31可采用Canny算子、拉普拉斯算子或拉普拉斯高斯(Laplacian of Gassian，LOG)算子提取图像的边缘特征；采用奇异值分解(Singular Value Decomposition，SVD)算法提取图像的纹理特征；采用方向梯度直方图(Histogram of Oriented Gradient，HOG)描述子或尺度不变特征转换(Scale-Invariant Feature Transform，SIFT)算法得到图像的特征向量；采用主成分分析(Principal Component Analysis，PCA)算法、线性判别分析(Linear Discriminant Analysis，LDA)算法或独立成分分析(Independent components analysis，ICA)算法等提取图像的全局或局部特征等。以达到去噪、提高识别效果的作用。例如，PCA是通过使用一个低维的特征向量来和投影矩阵来重建样本，通过最小化重建误差来对特征向量进行建模。同时，它是把方差大的维度留下，方差小的维度去掉。去掉方差小的维度可以帮助样本空间减小不确定性，留下方差大的维度可以保持样本和样本之间的局部距离。
具体实现中，所述自动聚类模块32可不需要任何先验知识，事先不了解各待识别的图像中的每一个图像的类别，而根据各待识别的图像的特征进行分类，将具有相似或相同特征的图像划分到同一子类。
具体实现中，当无法预先得知子类的个数，自动聚类模块32可采用基于概率分布模型的方法进行聚类分析，如狄利克雷过程聚类(Dirichlet Processes Clustering)算法；或先采用Canopy聚类算法进行预处理，再采用基于划分的方法进行聚类分析，如K-均值(K-Means)聚类算法。
具体实现中，所述自动聚类模块32还可可根据自动聚类结果，更新所述历史聚类结果。
具体实现中，如图4所示，所述第二分类模块34可进一步包括：
标注模块341，用于对于所述自动聚类结果中包括的未知的分类，进行类别 标注；
分类学习模块342，用于通过迁移学习，对所述带有类别标注的未知的分类训练分类器，以对所述图像中的未知目标物进行识别。
具体实现中，标注模块341具体可用于通过人机交互的方式，获取用户输入的类别标注信息；或者，从互联网查找与所述未知的分类相似性达到指定要求的图像，并通过互联网对所述图像的标注信息，对所述未知的分类进行类别标注。
比如，假设已知类别中，不包括“高山”这样的类别，则在步骤S104，对于自动聚类的结果，可通过用户给予“高山”的类别标注。或者，通过查找互联网，发现局聚类结果中的某一未知的分类的相似度很高的物体被称之为“高山”，则可采用互联网上的“高山”对所述未知的分类进行类别标注。
具体实现中，所述分类学习模块342具体可用于根据自动聚类的结果，对已有的分类器进行更新，并使用迁移学习方法和所述更新后的已有的分类器为所述带有类别标注的未知的分类训练分类器，以对所述图像中的未知目标物进行识别。在一些可行的实施方式中，已有和训练出的分类器可以是支持向量机(Support Vector Machine，SVM)分类器、贝叶斯分类器、决策树分类器、朴素贝叶斯分类器(Naive Bayes Classifier，NBC)等。
具体实现中，本发明实施例所述的迁移学习方法包括但不限于：covariance shift，TrAdaboost，基于多任务的学习等方法。
本发明实施例，当采用迁移学习方式后，其随着时间的增加，数据的增加，使学习的起点更高，收敛速度更快，训练出的分类器更优。并且，其可用新的聚类结果更新历史聚类，以及通过自动聚类结果更新已有的分类器，由此可实现整个系统的不断更新和不断学习，进而使系统不断优化。
由上可见，在本发明的一些可行的实施方式中，从图像中提取特征数据，并对所述提取的特征数据进行表达处理；根据所述表达处理后的特征数据和历史聚类结果，对图像中的特征进行自动聚类；对于所述自动聚类结果中包括的已知的分类，归类为对应的已知分类，以对所述图像中的已知目标物进行识别；对于所述自动聚类结果中包括的未知分类，通过机器学习的方式训练分类器，以对所述图像中的未知目标物进行识别。由此，本发明实施例可在待识别的图像中包含不属于已有类别的目标物时，对不属于已有类别的目标物进行学习，得到新的分类器，从而实现对未知分类的目标物进行识别。
图5为本发明的图像处理装置的另一实施例的结构组成示意图。如图5所述，其可包括：存储器51和处理器52，其中，处理器52调用存储器51中存储的程序代码，并执行如下步骤：
从图像中提取特征数据，并将所述提取的特征数据转换为统一的表达；
根据所述统一表达后的特征数据和历史聚类结果，对图像中的特征进行自动聚类；
对于所述自动聚类结果中包括的已知的分类，归类为对应的已知分类，以对所述图像中的已知目标物进行识别；
对于所述自动聚类结果中包括的未知分类，通过机器学习的方式训练分类器，以对所述图像中的未知目标物进行识别。
在一些可行的实施方式中，所述处理器执行对于所述自动聚类结果中包括的未知分类，通过机器学习的方式训练分类器，以对所述图像中的未知目标物进行识别的步骤时，可具体执行如下步骤：
对于所述自动聚类结果中包括的未知的分类，进行类别标注；
通过迁移学习，对所述带有类别标注的未知的分类训练分类器，以对所述 图像中的未知目标物进行识别。
在一些可行的实施方式中，所述处理器52进行类别标注，包括：
通过人机交互的方式，获取用户输入的类别标注信息；
或者，从互联网查找与所述未知的分类相似性达到指定要求的图像，并通过互联网对所述图像的标注信息，对所述未知的分类进行类别标注。
在一些可行的实施方式中，所述处理器52通过迁移学习，对所述带有类别标注的未知的分类训练分类器，以对所述图像中的未知目标物进行识别时，具体执行如下步骤：
根据自动聚类的结果，对已有的分类器进行更新；
使用迁移学习方法和所述更新后的已有的分类器为所述带有类别标注的未知的分类训练分类器，以对所述图像中的未知目标物进行识别。
在一些可行的实施方式中，当所述处理器52所述对图像中的特征进行自动聚类之后，还包括：
根据所述自动聚类结果，更新所述历史聚类结果。
本发明实施例的模块,可用通用集成电路(如中央处理器CPU),或以专用集成电路(ASIC)来实现。本领域普通技术人员可以理解上述实施例的各种方法中的全部或部分步骤是可以通过程序来指令相关的来完成，该程序可以存储于一计算机可读存储介质中，存储介质可以包括：闪存盘、只读存储器(Read-Only Memory，ROM)、随机存取器(Random Access Memory，RAM)、磁盘或光盘等。
以上所列举的仅为本发明较佳实施例而已，当然不能以此来限定本发明之权利范围，因此依本发明权利要求所作的等同变化，仍属本发明所涵盖的范围。

Claims (10)

  1. 一种图像中的目标物的识别方法，其特征在于，包括：
    从图像中提取特征数据，并将所述提取的特征数据转换为统一的表达；
    根据所述统一表达后的特征数据和历史聚类结果，对图像中的特征进行自动聚类；
    对于所述自动聚类结果中包括的已知的分类，归类为对应的已知分类，以对所述图像中的已知目标物进行识别；
    对于所述自动聚类结果中包括的未知分类，通过机器学习的方式训练分类器，以对所述图像中的未知目标物进行识别。
  2. 如权利要求1所述的图像中的目标物的识别方法，其特征在于，所述对于所述自动聚类结果中包括的未知分类，通过机器学习的方式训练分类器，以对所述图像中的未知目标物进行识别，包括：
    对于所述自动聚类结果中包括的未知的分类，进行类别标注；
    通过迁移学习，对所述带有类别标注的未知的分类训练分类器，以对所述图像中的未知目标物进行识别。
  3. 如权利要求2所述的图像中的目标物的识别方法，其特征在于，所述进行类别标注，包括：
    通过人机交互的方式，获取用户输入的类别标注信息；
    或者，从互联网查找与所述未知的分类相似性达到指定要求的图像，并通过互联网对所述图像的标注信息，对所述未知的分类进行类别标注。
  4. 如权利要求2所述的图像中的目标物的识别方法，其特征在于，通过迁移学习，对所述带有类别标注的未知的分类训练分类器，以对所述图像中的未知目标物进行识别，包括：
    根据自动聚类的结果，对已有的分类器进行更新；
    使用迁移学习方法和所述更新后的已有的分类器为所述带有类别标注的未知的分类训练分类器，以对所述图像中的未知目标物进行识别。
  5. 如权利要求1-4中任一项所述的图像中的目标物的识别方法，所述对图像中的特征进行自动聚类之后，还包括：
    根据所述自动聚类结果，更新所述历史聚类结果。
  6. 一种图像处理装置，其特征在于，包括：
    特征获取模块，用于从图像中提取特征数据，并将所述提取的特征数据转换为统一的表达；
    自动聚类模块，用于根据所述特征获取模块统一表达后的特征数据和历史聚类结果，对图像中的特征进行自动聚类；
    第一分类模块，用于对于所述自动聚类结果中包括的已知的分类，归类为对应的已知分类，以对所述图像中的已知目标物进行识别；
    第二分类模块，用于对于所述自动聚类结果中包括的未知分类，通过机器学习的方式训练分类器，以对所述图像中的未知目标物进行识别。
  7. 如权利要求6所述的图像处理装置，其特征在于，所述第二分类模块，包括：
    标注模块，用于对于所述自动聚类结果中包括的未知的分类，进行类别标注；
    分类学习模块，用于通过迁移学习，对所述带有类别标注的未知的分类训练分类器，以对所述图像中的未知目标物进行识别。
  8. 如权利要求7所述的图像处理装置，其特征在于，标注模块具体用于通过人机交互的方式，获取用户输入的类别标注信息；或者，从互联网查找与所述未知的分类相似性达到指定要求的图像，并通过互联网对所述图像的标注信息，对所述未知的分类进行类别标注。
  9. 如权利要求7所述的图像处理装置，其特征在于，所述分类学习模块具体用于根据自动聚类的结果，对已有的分类器进行更新，并使用迁移学习方法和所述更新后的已有的分类器为所述带有类别标注的未知的分类训练分类器， 以对所述图像中的未知目标物进行识别。
  10. 如权利要求6-9中任一项所述的图像处理装置，所述自动聚类模块还用于根据所述自动聚类结果，更新所述历史聚类结果。
PCT/CN2014/090976 2013-12-27 2014-11-13 图像中的目标物的识别方法及装置 WO2015096565A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US15/193,209 US9798956B2 (en) 2013-12-27 2016-06-27 Method for recognizing target object in image, and apparatus

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201310739555.3A CN104751198B (zh) 2013-12-27 2013-12-27 图像中的目标物的识别方法及装置
CN201310739555.3 2013-12-27

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US15/193,209 Continuation US9798956B2 (en) 2013-12-27 2016-06-27 Method for recognizing target object in image, and apparatus

Publications (1)

Publication Number Publication Date
WO2015096565A1 true WO2015096565A1 (zh) 2015-07-02

Family

ID=53477502

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2014/090976 WO2015096565A1 (zh) 2013-12-27 2014-11-13 图像中的目标物的识别方法及装置

Country Status (3)

Country Link
US (1) US9798956B2 (zh)
CN (1) CN104751198B (zh)
WO (1) WO2015096565A1 (zh)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111931670A (zh) * 2020-08-14 2020-11-13 成都数城科技有限公司 基于卷积神经网的深度图像头部检测与定位方法及系统
CN116030388A (zh) * 2022-12-30 2023-04-28 以萨技术股份有限公司 一种识别任务的处理方法、电子设备及存储介质

Families Citing this family (36)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104751198B (zh) * 2013-12-27 2018-04-27 华为技术有限公司 图像中的目标物的识别方法及装置
CN108460389B (zh) 2017-02-20 2021-12-03 阿里巴巴集团控股有限公司 一种识别图像中对象的类型预测方法、装置及电子设备
CN106951899A (zh) * 2017-02-24 2017-07-14 李刚毅 基于图像识别的异常检测方法
CN107239514A (zh) * 2017-05-19 2017-10-10 邓昌顺 一种基于卷积神经网络的植物识别方法及系统
US10783393B2 (en) 2017-06-20 2020-09-22 Nvidia Corporation Semi-supervised learning for landmark localization
CN111417961B (zh) * 2017-07-14 2024-01-12 纪念斯隆-凯特林癌症中心 弱监督的图像分类器
CN108319890A (zh) * 2017-12-01 2018-07-24 中国电子科技集团公司电子科学研究院 基于多视角图像处理的指静脉识别方法、设备及存储介质
CN108256550A (zh) * 2017-12-14 2018-07-06 北京木业邦科技有限公司 一种木材类别更新方法和装置
CN109977975B (zh) * 2017-12-28 2022-11-22 沈阳新松机器人自动化股份有限公司 物品回收系统和物品回收方法
CN108376235A (zh) * 2018-01-15 2018-08-07 深圳市易成自动驾驶技术有限公司 图像检测方法、装置及计算机可读存储介质
CN108734182B (zh) * 2018-06-13 2022-04-05 大连海事大学 一种基于小数据样本学习的兴趣特征识别检测方法
CN108985214A (zh) * 2018-07-09 2018-12-11 上海斐讯数据通信技术有限公司 图像数据的标注方法和装置
CN110712368A (zh) * 2018-07-13 2020-01-21 三纬国际立体列印科技股份有限公司 整合式3d打印系统
CN110751673B (zh) * 2018-07-23 2022-08-19 中国科学院长春光学精密机械与物理研究所 一种基于集成学习的目标跟踪方法
KR102629036B1 (ko) * 2018-08-30 2024-01-25 삼성전자주식회사 로봇 및 그의 제어 방법
CN109376764B (zh) * 2018-09-13 2021-12-07 北京字节跳动网络技术有限公司 基于聚类的数据收集方法、装置和计算机可读存储介质
KR102533148B1 (ko) 2018-09-17 2023-05-17 데이터로그, 엘엘씨 통나무 검척 시스템 및 관련 방법
CN109344890A (zh) * 2018-09-20 2019-02-15 浪潮软件股份有限公司 一种基于深度学习的烟柜卷烟识别方法
CN110046632B (zh) * 2018-11-09 2023-06-02 创新先进技术有限公司 模型训练方法和装置
CN109800781A (zh) * 2018-12-07 2019-05-24 北京奇艺世纪科技有限公司 一种图像处理方法、装置及计算机可读存储介质
US10832096B2 (en) * 2019-01-07 2020-11-10 International Business Machines Corporation Representative-based metric learning for classification and few-shot object detection
CN109992682A (zh) * 2019-03-29 2019-07-09 联想(北京)有限公司 一种图像识别方法、装置及电子设备
CN111796663B (zh) * 2019-04-09 2022-08-16 Oppo广东移动通信有限公司 场景识别模型更新方法、装置、存储介质及电子设备
CN110147824B (zh) * 2019-04-18 2021-04-02 微梦创科网络科技（中国）有限公司 一种图像的自动分类方法及装置
CN110298839A (zh) * 2019-07-10 2019-10-01 北京滴普科技有限公司 一种基于数据驱动的外观缺陷智能判别系统
CN110598636B (zh) * 2019-09-09 2023-01-17 哈尔滨工业大学 一种基于特征迁移的舰船目标识别方法
US11900679B2 (en) * 2019-11-26 2024-02-13 Objectvideo Labs, Llc Image-based abnormal event detection
CN111191690B (zh) * 2019-12-16 2023-09-05 上海航天控制技术研究所 基于迁移学习的空间目标自主识别方法、电子设备和存储介质
CN111339343A (zh) * 2020-02-12 2020-06-26 腾讯科技（深圳）有限公司 图像检索方法、装置、存储介质及设备
CN111652292B (zh) * 2020-05-20 2022-12-06 贵州电网有限责任公司 一种基于ncs、ms的相似物体实时检测方法及系统
CN111860606B (zh) * 2020-06-24 2021-09-14 上海小零网络科技有限公司 图像分类的方法、装置以及存储介质
CN112364899A (zh) * 2020-10-27 2021-02-12 西安科技大学 一种基于虚拟图像与迁移学习的磨粒铁谱图像智能识别方法
CN114648480A (zh) * 2020-12-17 2022-06-21 杭州海康威视数字技术股份有限公司 表面缺陷检测方法、装置及系统
CN113743443B (zh) * 2021-05-31 2024-05-17 高新兴科技集团股份有限公司 一种图像证据分类和识别方法及装置
CN113673383B (zh) * 2021-08-05 2024-04-19 苏州智加科技有限公司 一种面向复杂道路场景的时空域障碍物检测方法及系统
CN116152721B (zh) * 2023-04-18 2023-06-20 北京航空航天大学 一种基于退火式标签迁移学习的目标检测方法及装置

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1559051A (zh) * 2001-09-28 2004-12-29 �ʼҷ����ֵ��ӹɷ����޹�˾ 使用部分学习模型的面部识别的系统和方法
CN101226590A (zh) * 2008-01-31 2008-07-23 湖南创合制造有限公司 一种人脸识别方法
US20110158535A1 (en) * 2009-12-24 2011-06-30 Canon Kabushiki Kaisha Image processing apparatus and image processing method
CN102289686A (zh) * 2011-08-09 2011-12-21 北京航空航天大学 一种基于迁移学习的运动目标分类识别方法
CN103177264A (zh) * 2013-03-14 2013-06-26 中国科学院自动化研究所 基于视觉词典全局拓扑表达的图像分类方法

Family Cites Families (13)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7298903B2 (en) * 2001-06-28 2007-11-20 Microsoft Corporation Method and system for separating text and drawings in digital ink
US7233692B2 (en) * 2002-11-14 2007-06-19 Lockheed Martin Corporation Method and computer program product for identifying output classes with multi-modal dispersion in feature space and incorporating multi-modal structure into a pattern recognition system
JP2013092911A (ja) * 2011-10-26 2013-05-16 Sony Corp 情報処理装置、情報処理方法、および、プログラム
CN102546625A (zh) * 2011-12-31 2012-07-04 深圳市永达电子股份有限公司 半监督聚类集成的协议识别系统
US8811727B2 (en) * 2012-06-15 2014-08-19 Moataz A. Rashad Mohamed Methods for efficient classifier training for accurate object recognition in images and video
CN103065122A (zh) * 2012-12-21 2013-04-24 西北工业大学 基于面部动作单元组合特征的人脸表情识别方法
CN103117903B (zh) * 2013-02-07 2016-01-06 中国联合网络通信集团有限公司 上网流量异常检测方法及装置
US10169686B2 (en) * 2013-08-05 2019-01-01 Facebook, Inc. Systems and methods for image classification by correlating contextual cues with images
KR102190484B1 (ko) * 2013-11-11 2020-12-11 삼성전자주식회사 인식기 학습 방법 및 장치, 데이터 인식 방법 및 장치
CN104751198B (zh) * 2013-12-27 2018-04-27 华为技术有限公司 图像中的目标物的识别方法及装置
US9767386B2 (en) * 2015-06-23 2017-09-19 Adobe Systems Incorporated Training a classifier algorithm used for automatically generating tags to be applied to images
US9704054B1 (en) * 2015-09-30 2017-07-11 Amazon Technologies, Inc. Cluster-trained machine learning for image processing
US10068129B2 (en) * 2015-11-18 2018-09-04 Adobe Systems Incorporated Recognizing unknown person instances in an image gallery

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1559051A (zh) * 2001-09-28 2004-12-29 �ʼҷ����ֵ��ӹɷ����޹�˾ 使用部分学习模型的面部识别的系统和方法
CN101226590A (zh) * 2008-01-31 2008-07-23 湖南创合制造有限公司 一种人脸识别方法
US20110158535A1 (en) * 2009-12-24 2011-06-30 Canon Kabushiki Kaisha Image processing apparatus and image processing method
CN102289686A (zh) * 2011-08-09 2011-12-21 北京航空航天大学 一种基于迁移学习的运动目标分类识别方法
CN103177264A (zh) * 2013-03-14 2013-06-26 中国科学院自动化研究所 基于视觉词典全局拓扑表达的图像分类方法

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111931670A (zh) * 2020-08-14 2020-11-13 成都数城科技有限公司 基于卷积神经网的深度图像头部检测与定位方法及系统
CN111931670B (zh) * 2020-08-14 2024-05-31 成都数城科技有限公司 基于卷积神经网的深度图像头部检测与定位方法及系统
CN116030388A (zh) * 2022-12-30 2023-04-28 以萨技术股份有限公司 一种识别任务的处理方法、电子设备及存储介质
CN116030388B (zh) * 2022-12-30 2023-08-11 以萨技术股份有限公司 一种识别任务的处理方法、电子设备及存储介质

Also Published As

Publication number Publication date
US9798956B2 (en) 2017-10-24
CN104751198A (zh) 2015-07-01
US20160307070A1 (en) 2016-10-20
CN104751198B (zh) 2018-04-27

Similar Documents

Publication Publication Date Title
WO2015096565A1 (zh) 图像中的目标物的识别方法及装置
Li et al. Joint visual and temporal consistency for unsupervised domain adaptive person re-identification
Chuang et al. A feature learning and object recognition framework for underwater fish images
Li et al. Efficient boosted exemplar-based face detection
US9317781B2 (en) Multiple cluster instance learning for image classification
Charles et al. Automatic and efficient human pose estimation for sign language videos
CN108520226B (zh) 一种基于躯体分解和显著性检测的行人重识别方法
EP3002710A1 (en) System and method for object re-identification
Long et al. Accurate object detection with location relaxation and regionlets re-localization
Bhunia et al. Text recognition in scene image and video frame using color channel selection
US20150110387A1 (en) Method for binary classification of a query image
EP2409250A1 (en) Semantic event detection using cross-domain knowledge
CN103824052A (zh) 一种基于多层次语义特征的人脸特征提取方法及识别方法
US20100111375A1 (en) Method for Determining Atributes of Faces in Images
Nag et al. A new unified method for detecting text from marathon runners and sports players in video (PR-D-19-01078R2)
Zou et al. Online glocal transfer for automatic figure-ground segmentation
US20240087352A1 (en) System for identifying companion animal and method therefor
CN115497124A (zh) 身份识别方法和装置及存储介质
CN115203408A (zh) 一种多模态试验数据智能标注方法
Wang et al. Action recognition based on object tracking and dense trajectories
Wang et al. Face tracking and recognition via incremental local sparse representation
Almomani et al. Object tracking via Dirichlet process-based appearance models
Yan et al. Real time lobster posture estimation for behavior research
Shen et al. Novel text recognition based on modified k-clustering and hidden markov models
Ghavidel et al. Natural scene text localization using edge color signature

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: 14875767

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 14875767

Country of ref document: EP

Kind code of ref document: A1