WO2019080430A1 - 电子装置、混乱样本整理方法和计算机可读存储介质 - Google Patents
电子装置、混乱样本整理方法和计算机可读存储介质Info
- Publication number
- WO2019080430A1 WO2019080430A1 PCT/CN2018/077671 CN2018077671W WO2019080430A1 WO 2019080430 A1 WO2019080430 A1 WO 2019080430A1 CN 2018077671 W CN2018077671 W CN 2018077671W WO 2019080430 A1 WO2019080430 A1 WO 2019080430A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- sample
- local
- feature
- feature set
- point
- 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
- 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
- G06V10/20—Image preprocessing
- G06V10/32—Normalisation of the pattern dimensions
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
Definitions
- the present application relates to the field of sample data processing, and in particular, to an electronic device, a chaotic sample sorting method, and a computer readable storage medium.
- the present application provides an electronic device, a chaotic sample sorting method, and a computer readable storage medium, which are intended to reduce the time consuming of sample data collation and improve the collation efficiency.
- a first aspect of the present application provides an electronic device including a memory and a processor, the memory storing a chaotic sample organizing system executable on the processor, when the chaotic sample organizing system is executed by the processor Implement the following steps:
- the local features in the second feature set are clustered into pre-set n categories by using a preset type of clustering algorithm, and all the sample pictures obtained are divided into n categories according to the local feature clustering results in the second feature set.
- the second aspect of the present application provides a chaotic sample sorting method, the method comprising the steps of:
- the local features in the second feature set are clustered into preset n categories by using a preset type of clustering algorithm, and all the sample pictures obtained are classified into n types according to the local feature clustering result in the second feature set. .
- a third aspect of the present application provides a computer readable storage medium storing a chaotic sample sorting system, the chaotic sample sorting method system being executable by at least one processor to cause the at least one processing Perform the following steps:
- the local features in the second feature set are clustered into preset n categories by using a preset type of clustering algorithm, and all the sample pictures obtained are classified into n types according to the local feature clustering result in the second feature set. .
- the scheme automatically sorts and samples the sample images through the system, and the sample sorting speed is fast and the efficiency is high, which solves the problem that the manual sorting of the samples takes too long and affects the progress of the sample training work, and reduces the problem. Labor cost.
- FIG. 1 is a schematic flow chart of an embodiment of a messy sample sorting method according to the present application.
- FIG. 2 is a schematic flow chart of a second embodiment of a chaotic sample sorting method according to the present application.
- FIG. 3 is a schematic diagram of an operating environment of an embodiment of a chaotic sample organizing system of the present application.
- FIG. 4 is a block diagram of a program of an embodiment of a chaotic sample organizing system of the present application.
- FIG. 5 is a program block diagram of a second embodiment of the chaotic sample organizing system of the present application.
- FIG. 1 is a schematic flowchart of an embodiment of a chaotic sample sorting method according to the present application.
- the chaotic sample sorting method includes:
- Step S10 Obtain a sample image to be sorted, and uniformly adjust the obtained sample image to a preset size
- the system obtains the sample image to be collated from the sample library, and the size of the sample image obtained may be inconsistent. Therefore, the size of the sample image is first adjusted, and the size of all sample images is uniformly adjusted to a preset size. Subsequent calculations are prepared; the preset size should be smaller than the size of the smallest sample picture in all sample pictures. For example, sample images of various sizes such as 800 pixels ⁇ 600 pixels and 480 pixels ⁇ 800 pixels are present in the acquired sample picture, and the size of the sample pictures is uniformly adjusted to 256 pixels ⁇ 256 pixels (ie, a preset size).
- Step S20 performing local feature extraction on all sample images after resizing to obtain a first feature set
- the system After adjusting the size of the sample image, the system performs local feature extraction for each sample image, and extracts local features of all sample images to form a first feature set.
- the local features may be LBP (Local binary patterns) features, HOG (Histogram of Oriented Gradient) features, SURF (Speeded Up Robust Features) features, and the like.
- the local feature of the sample picture may be extracted as follows: 1. For each sample picture, extract a local feature; 2.
- the local feature is used as a local feature of the sample picture, such that the local feature of the sample picture can better represent the sample picture.
- Step S30 performing PCA dimensionality reduction processing on the local features in the first feature set, and obtaining a second feature set after the dimension reduction processing;
- All local features in the first feature set are subjected to dimensionality reduction processing by a PCA (Principal Component Analysis) algorithm, and all local features after dimensionality reduction form a second feature set.
- the local feature extracted from the sample image is a high-dimensional vector composed of a series of features (for example, the dimension reaches 5000), wherein the elements of some dimensions are equal in all sample images or have little difference from each other, then these dimensional elements are The contribution of distinguishing sample images is small; the purpose of dimension reduction by PCA is to remove the dimension elements in the first feature set that contribute little to distinguishing sample images, and to find those dimensional elements with large variance in the first feature set. (Dimension elements that contribute a lot to the sample image); thus the dimension elements of each local feature.
- Step S40 The local feature in the second feature set is clustered into a preset n class by using a preset type of clustering algorithm, and all the sample images obtained are divided according to the local feature clustering result in the second feature set. n class.
- the clustering result of the local features in the feature set is divided into n categories corresponding to the sample pictures to which the local features belong, so that the classification of the sample pictures to be processed is completed.
- the sample pictures are uniformly sized and adjusted, and then the local features of each sample picture after the resizing are extracted to form a first feature set, and then the local features in the first feature combination are subjected to PCA drop.
- Dimensional processing the second feature set formed by the local feature after the dimension reduction processing is obtained, and finally the local features in the second feature set are clustered by a preset clustering algorithm, and the local features are clustered into preset n classes.
- the corresponding sample pictures are sorted into n categories, that is, the sorting and sorting of the sample pictures is completed.
- the scheme automatically sorts and samples the sample images through the system, and the sample sorting speed is fast and the efficiency is high, which solves the problem that the manual sorting of the samples takes too long and affects the progress of the sample training work, and reduces the problem. Labor cost.
- FIG. 2 is a schematic flowchart of the second embodiment of the method for confusing sample preparation according to the present application.
- the solution of the embodiment is based on an embodiment.
- the method for confusing the chaotic sample is before the step S20.
- Step S50 Calculate the obtained number of sample pictures, determine the number of folders corresponding to the current sample picture number according to the mapping relationship between the preset sample quantity interval and the number of folders, and evenly distribute all the sample pictures after the resizing to the determined quantity.
- the folder
- the present embodiment divides the sample picture into a plurality of folders, and performs subsequent calculation and processing in units of folders.
- the system pre-sets a mapping relationship between the sample quantity interval and the number of allocated folders. After calculating the number of sample images acquired, the system can determine the sample image by querying the mapping relationship table according to the sample number interval to which the sample picture quantity belongs.
- the number of folders to be allocated corresponding to the quantity after determining the number of folders, all the sample images that have been uniformly adjusted in size are evenly distributed to a certain number of folders.
- sample images there are 100,000 sample images, and the number of folders corresponding to the sample size range of 9 to 110,000 is 10, and the 100,000 sample images are opened to 10 folders of folder 1, folder 2, ... folder10, each Put 10,000 sheets in a folder.
- sample images may be allocated to various folders in other manners.
- the sample picture may not be divided into folders, or the number of sample pictures corresponds to one folder, that is, the sample picture does not need to be allocated.
- the chaotic sample sorting method of the embodiment replaces the step S40 with:
- Step S60 The local feature in the second feature set is clustered into a preset n class by using a preset type clustering algorithm, and the sample in each folder is obtained according to the local feature clustering result in the second feature set.
- the picture is divided into n categories.
- the preset type of clustering algorithm used is a k-means algorithm
- the step of clustering the local features in the second feature set into the preset n categories includes:
- n seed points eg, Euclidean distance, cosine distance
- Point group for example, the local feature Pi is closest to the seed point Si, then Pi is divided into Si point groups), and n point groups are obtained;
- each point in the point group represents a local feature of a sample picture.
- the present application also proposes a chaotic sample sorting system.
- FIG. 3 is a schematic diagram of an operating environment of a preferred embodiment of the chaotic sample organizing system 10 of the present application.
- the chaotic sample organizing system 10 is installed and operated in the electronic device 1.
- the electronic device 1 may be a computing device such as a desktop computer, a notebook, a palmtop computer, and a server.
- the electronic device 1 may include, but is not limited to, a memory 11, a processor 12, and a display 13.
- Figure 3 shows only the electronic device 1 with components 11-13, but it should be understood that not all illustrated components may be implemented, and more or fewer components may be implemented instead.
- the memory 11 may be an internal storage unit of the electronic device 1 in some embodiments, such as a hard disk or memory of the electronic device 1.
- the memory 11 may also be an external storage device of the electronic device 1 in other embodiments, such as a plug-in hard disk equipped on the electronic device 1, a smart memory card (SMC), and a secure digital (SD). Card, flash card, etc.
- the memory 11 may also include both an internal storage unit of the electronic device 1 and an external storage device.
- the memory 11 is used to store application software and various types of data installed in the electronic device 1, such as program code of the chaotic sample sorting system 10.
- the memory 11 can also be used to temporarily store data that has been output or is about to be output.
- the processor 12 in some embodiments, may be a Central Processing Unit (CPU), microprocessor or other data processing chip for running program code or processing data stored in the memory 11, such as performing chaotic sample processing. System 10 and so on.
- CPU Central Processing Unit
- microprocessor or other data processing chip for running program code or processing data stored in the memory 11, such as performing chaotic sample processing. System 10 and so on.
- the display 13 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, an OLED (Organic Light-Emitting Diode) touch sensor, or the like in some embodiments.
- the display 13 is for displaying information processed in the electronic device 1 and a user interface for displaying visualization, such as a business customization interface or the like.
- the components 11-13 of the electronic device 1 communicate with one another via a system bus.
- FIG. 4 is a program module diagram of an embodiment of the chaotic sample organizing system 10 of the present application.
- the chaotic sample organization system 10 can be divided into one or more modules, one or more modules are stored in the memory 11, and by one or more processors (the processor 12 in this embodiment) Executed to complete the application.
- the chaotic sample organization system 10 can be divided into an acquisition module 101, an extraction module 102, a dimensionality reduction module 103, and a first clustering module 104.
- a module referred to in this application refers to a series of computer program instruction segments capable of performing a specific function, and is more suitable than the program for describing the execution process of the chaotic sample organization system 10 in the electronic device 1, wherein:
- the obtaining module 101 is configured to obtain a sample image to be collated, and uniformly adjust the obtained sample image to a preset size;
- the system obtains the sample image to be collated from the sample library, and the size of the sample image obtained may be inconsistent. Therefore, the size of the sample image is first adjusted, and the size of all sample images is uniformly adjusted to a preset size. Subsequent calculations are prepared; the preset size should be smaller than the size of the smallest sample picture in all sample pictures. For example, sample images of various sizes such as 800 pixels ⁇ 600 pixels and 480 pixels ⁇ 800 pixels are present in the acquired sample picture, and the size of the sample pictures is uniformly adjusted to 256 pixels ⁇ 256 pixels (ie, a preset size).
- the extracting module 102 is configured to perform local feature extraction on all the sample images after the resizing to obtain the first feature set;
- the system After adjusting the size of the sample image, the system performs local feature extraction for each sample image, and extracts local features of all sample images to form a first feature set.
- the local features may be LBP (Local binary patterns) features, HOG (Histogram of Oriented Gradient) features, SURF (Speeded Up Robust Features) features, and the like.
- the local feature of the sample picture may be extracted as follows: 1. For each sample picture, extract a local feature; 2.
- the local feature is used as a local feature of the sample picture, such that the local feature of the sample picture can better represent the sample picture.
- the dimension reduction module 103 is configured to perform a PCA dimensionality reduction process on the local features in the first feature set to obtain a second feature set after the dimension reduction process;
- All local features in the first feature set are subjected to dimensionality reduction processing by a PCA (Principal Component Analysis) algorithm, and all local features after dimensionality reduction form a second feature set.
- the local feature extracted from the sample image is a high-dimensional vector composed of a series of features (for example, the dimension reaches 5000), wherein the elements of some dimensions are equal in all sample images or have little difference from each other, then these dimensional elements are The contribution of distinguishing sample images is small; the purpose of dimension reduction by PCA is to remove the dimension elements in the first feature set that contribute little to distinguishing sample images, and to find those dimensional elements with large variance in the first feature set. (Dimension elements that contribute a lot to the sample image).
- the first clustering module 104 is configured to cluster the local features in the second feature set into a preset n class by using a preset type of clustering algorithm, and obtain the local feature clustering result according to the second feature set. All sample images arriving are divided into n categories.
- the clustering result of the local features in the feature set is divided into n categories corresponding to the sample pictures to which the local features belong, so that the classification of the sample pictures to be processed is completed.
- the sample pictures are uniformly sized and adjusted, and then the local features of each sample picture after the resizing are extracted to form a first feature set, and then the local features in the first feature combination are subjected to PCA drop.
- Dimensional processing the second feature set formed by the local feature after the dimension reduction processing is obtained, and finally the local features in the second feature set are clustered by a preset clustering algorithm, and the local features are clustered into preset n classes.
- the corresponding sample pictures are sorted into n categories, that is, the sorting and sorting of the sample pictures is completed.
- the scheme automatically sorts and samples the sample images through the system, and the sample sorting speed is fast and the efficiency is high, which solves the problem that the manual sorting of the samples takes too long and affects the progress of the sample training work, and reduces the problem. Labor cost.
- the chaotic sample organization system further includes:
- the sample allocation module 105 is configured to calculate the obtained number of sample pictures, determine the number of folders corresponding to the current sample picture number according to a preset mapping relationship between the sample number interval and the number of folders, and evenly adjust all the sample pictures after the size adjustment. Assigned to a defined number of folders;
- the present embodiment divides the sample picture into a plurality of folders, and performs subsequent calculation and processing in units of folders.
- the system pre-sets a mapping relationship between the sample quantity interval and the number of allocated folders. After calculating the number of sample images acquired, the system can determine the sample image by querying the mapping relationship table according to the sample number interval to which the sample picture quantity belongs.
- the number of folders to be allocated corresponding to the quantity after determining the number of folders, all the sample images that have been uniformly adjusted in size are evenly distributed to a certain number of folders.
- sample images there are 100,000 sample images, and the number of folders corresponding to the sample size range of 9 to 110,000 is 10, and the 100,000 sample images are opened to 10 folders of folder 1, folder 2, ... folder10, each Put 10,000 sheets in a folder.
- sample images may be allocated to various folders in other manners.
- the sample picture may not be divided into folders, or the number of sample pictures corresponds to one folder, that is, the sample picture does not need to be allocated.
- the first clustering module 104 is replaced with a second clustering module 106, and the second clustering module 106 is configured to cluster the local features in the second feature set into a preset n by using a preset type of clustering algorithm.
- Class according to the results of the cluster, the sample pictures in each folder are divided into n categories.
- the preset type of clustering algorithm adopted by the first clustering module 104 and the second clustering module 106 in the embodiment is a k-means algorithm, and the local features in the second feature set are clustered into pre-
- the steps for the n categories are as follows:
- n seed points eg, Euclidean distance, cosine distance
- Point group for example, the local feature Pi is closest to the seed point Si, then Pi is divided into Si point groups), and n point groups are obtained;
- each point in the point group represents a local feature of a sample picture.
- the present application further provides a computer readable storage medium storing a chaotic sample organizing system, the chaotic sample organizing system being executable by at least one processor to cause the at least one processing
- the chaotic sample arranging method in any of the above embodiments is performed.
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Computational Biology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Image Analysis (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
一种电子装置、混乱样本整理方法和计算机可读存储介质,其中,该方法包括:获取待整理的样本图片,将获取的样本图片统一调整至预设尺寸(S10);对调整尺寸后的所有样本图片进行局部特征提取,以得到第一特征集合(S20);对所述第一特征集合中的局部特征进行PCA降维处理,得到降维处理后的第二特征集合(S30);采用预设类型的聚类算法将第二特征集合中的局部特征聚类成预先设置的n类,根据第二特征集合中的局部特征聚类结果,将获取到的所有样本图片分成n类(S40)。上述技术方案有效提升了样本整理效率高,解决了人工整理样本耗时太长而影响样本训练工作的进度的问题,并且降低了人工成本。
Description
本申请基于巴黎公约申明享有2017年10月27日递交的申请号为CN 201711025204.0、名称为“电子装置、混乱样本整理方法和计算机可读存储介质”中国专利申请的优先权,该中国专利申请的整体内容以参考的方式结合在本申请中。
本申请涉及样本数据处理领域,特别涉及一种电子装置、混乱样本整理方法和计算机可读存储介质。
使用深度学习进行图片样本数据训练时,需要大量的样本数据,目前业内一般都是采用人工对样本数据进行整理分类。然而,当样本数据量巨大,达到数以万、几十万计时,通过人工对样本进行整理则极为耗时且效率低,严重影响样本训练工作的进度,并且人工成本非常高。
发明内容
本申请提供一种电子装置、混乱样本整理方法和计算机可读存储介质,旨在降低样本数据整理的耗时,提升整理效率。
本申请第一方面提供一种电子装置,包括存储器和处理器,所述存储器上存储有可在所述处理器上运行的混乱样本整理系统,所述混乱样本整理系统被所述处理器执行时实现如下步骤:
获取待整理的样本图片,将获取的样本图片统一调整至预设尺寸;
对调整尺寸后的所有样本图片进行局部特征提取,以得到第一特征集合;
对所述第一特征集合中的局部特征进行PCA降维处理,得到降维处理后的第二特征集合;
采用预设类型的聚类算法将第二特征集合中的局部特征聚类成预先设置的n类,根据第二特征集合中的局部特征聚类结果,将获取到的所有样本图片分成n类。
本申请第二方面提供一种混乱样本整理方法,该方法包括步骤:
获取待整理的样本图片,将获取的样本图片统一调整至预设尺寸;
对调整尺寸后的所有样本图片进行局部特征提取,以得到第一特征集合;
对所述第一特征集合中的局部特征进行PCA降维处理,得到第二特征集合;
采用预设类型的聚类算法将第二特征集合中的局部特征聚类成预设的n个类别,根据第二特征集合中的局部特征聚类结果,将获取到的所有样本图片分成n类。
本申请第三方面提供一种计算机可读存储介质,所述计算机可读存储介质存储有混乱样本整理系统,所述混乱样本整理方法系统可被至少一个处理器执行,以使所述至少一个处理器执行如下步骤:
获取待整理的样本图片,将获取的样本图片统一调整至预设尺寸;
对调整尺寸后的所有样本图片进行局部特征提取,以得到第一特征集合;
对所述第一特征集合中的局部特征进行PCA降维处理,得到第二特征集合;
采用预设类型的聚类算法将第二特征集合中的局部特征聚类成预设的n个类别,根据第二特征集合中的局部特征聚类结果,将获取到的所有样本图片分成n类。
本申请技术方案,首先对所有样本图片进行尺寸统一规范调整,然后提取调整尺寸后的每张样本图片的局部特征,组成第一特征集合,再将第一特征结合中的局部特征进行PCA降维处理,得到降维处理后的局部特征形成的第二特征集合,最终通过预设的聚类算法对第二特征集合中的局部特征进行聚类,将局部特征聚类成预设的n类,根据局部特征的分类,对应的将样本图片进行分类整理为n类,即完成样本图片的整理分类。与现有技术相比,本方案通过系统对样本图片进行自动的分类整理,样本整理速度快、效率高,解决了人工整理 样本耗时太长而影响样本训练工作的进度的问题,并且降低了人工成本。
图1为本申请混乱样本整理方法一实施例的流程示意图;
图2为本申请混乱样本整理方法二实施例的流程示意图;
图3为本申请混乱样本整理系统一实施例的运行环境示意图;
图4为本申请混乱样本整理系统一实施例的程序模块图;
图5为本申请混乱样本整理系统二实施例的程序模块图。
以下结合附图对本申请的原理和特征进行描述,所举实例只用于解释本申请,并非用于限定本申请的范围。
如图1所示,图1为本申请混乱样本整理方法一实施例的流程示意图。
本实施例中,该混乱样本整理方法包括:
步骤S10,获取待整理的样本图片,将获取的样本图片统一调整至预设尺寸;
首先,系统从样本库中获取待整理的样本图片,获取到的样本图片的尺寸大小可能不一致,因此先对获取的样本图片进行尺寸调整,将所有样本图片的尺寸统一调整至预设尺寸,为后续计算作准备;所述预设尺寸应该小于所有样本图片中的最小样本图片的尺寸。例如,获取的样本图片中存在800像素×600像素、480像素×800像素等各种尺寸的样本图片,统一将这些样本图片的尺寸调整为256像素×256像素(即预设尺寸)。
步骤S20,对调整尺寸后的所有样本图片进行局部特征提取,以得到第一特征集合;
系统在调整好样本图片的尺寸后,对每一张样本图片分别进行局部特征提取,提取出所有样本图片的局部特征形成第一特征集合。本实施例中,局部特征可以为LBP(Local binary patterns)特征、HOG(Histogram of Oriented Gradient)特征、SURF(Speeded Up Robust Features)特征,等等。优选地,本实施例中,对样本图片的局部特征的提取方式可以为:1、针对每一张样本图片,提取一个局部特征;2、针对每一张样本图片,先将该样本图片划分为预设的m(例如m=4)个图像区域,然后从各个图像区域中分别提取一个局部特征,再计算出从这m个图像区域提取出的m个局部特征的平均局部特征,将该平均局部特征作为该样本图片的局部特征,这样该样本图片的局部特征可以更好的代表该样本图片。
步骤S30,对所述第一特征集合中的局部特征进行PCA降维处理,得到降维处理后的第二特征集合;
对第一特征集合中的所有局部特征均通过PCA(Principal Component Analysis,主成分分析)算法进行降维处理,降维后的所有局部特征形成第二特征集合。从样本图片提取的局部特征是由一系列特征组成的高维向量(例如,维度达到5000),其中,某些维度的元素在所有样本图片中都相等或者彼此差距不大,那么这些维度元素对于区分样本图片的贡献度就很小;通过PCA降维的目的就是为去掉第一特征集合中这些对区分样本图片贡献度很小的维度元素,并找到第一特征集合中那些方差大的维度元素(对区分样本图片贡献度大的维度元素);从而使各个局部特征的维度元素。
步骤S40,采用预设类型的聚类算法将第二特征集合中的局部特征聚类成预先设置的n类,根据第二特征集合中的局部特征聚类结果,将获取到的所有样本图片分成n类。
采用预设类型的聚类算法对第二特征集合中的局部特征进行聚类,以将第二特征集合中的局部特征聚类成预先设置的n(例如n=3)类;然后根据第二特征集合中的局部特征的聚类结果,将各个局部特征所属的样本图片对应的分成n类,如此则完成对待处理的样本图片的分类整理。
本实施例技术方案,首先对所有样本图片进行尺寸统一规范调整,然后提取调整尺寸后的每张样本图片的局部特征,组成第一特征集合,再将第一特征结合中的局部特征进行PCA降维处理,得到降维处理后的局部特征形成的第二特征集合,最终通过预设的聚类算法 对第二特征集合中的局部特征进行聚类,将局部特征聚类成预设的n类,根据局部特征的分类,对应的将样本图片进行分类整理为n类,即完成样本图片的整理分类。与现有技术相比,本方案通过系统对样本图片进行自动的分类整理,样本整理速度快、效率高,解决了人工整理样本耗时太长而影响样本训练工作的进度的问题,并且降低了人工成本。
如图2所示,图2为本申请混乱样本整理方法二实施例的流程示意图;本实施例方案基于一实施例,在本实施例中,该混乱样本整理方法在所述步骤S20之前,还包括:
步骤S50,计算获取到的样本图片数量,根据预设的样本数量区间与文件夹数量的映射关系,确定当前样本图片数量对应的文件夹数量,将调整尺寸后的所有样本图片均匀分配至确定数量的文件夹中;
当样本图片数量巨大(例如样本图片多至几万或几十万张)时,为了减少计算量,本实施例将样本图片划分至多个文件夹,以文件夹为单位进行后续的计算、处理。系统预先设置了样本数量区间与分配文件夹数量的映射关系表,系统在计算出获取的样本图片数量后,根据该样本图片数量所属的样本数量区间,通过查询映射关系表就可确定该样本图片数量对应的需分配的文件夹数量,在确定文件夹数量后,将经过尺寸统一调整后的所有样本图片以均匀的方式分配至确定数量的文件夹中。比如有10万张样本图片,9至11万的样本数量区间对应的文件夹数量为10个,则将这10万张样本图片分开放至folder1、folder2……folder10这10个文件夹中,每个文件夹中放1万张。需要说明的是,本实施例仅仅是优选均匀分配方式为例,在其它实施例中,还可以采取其它方式将样本图片分配至各个文件夹中。另外,当样本图片数量不是很巨大(例如样本图片仅有几千张)时,可以不对样本图片进行文件夹的划分,或该样本图片数量对应为一个文件夹,即无需对样本图片进行分配。
并且,本实施例的混乱样本整理方法将所述步骤S40替换为:
步骤S60,采用预设类型的聚类算法将第二特征集合中的局部特 征聚类成预先设置的n类,根据第二特征集合中的局部特征聚类结果,将每个文件夹中的样本图片分成n类。
优选地,所述步骤S40中,采用的预设类型的聚类算法为k-means算法,所述将第二特征集合中的局部特征聚类成预设的n个类别的步骤包括:
a、随机在所述第二特征集合中选取n(例如n=3)个局部特征作为种子点;
b、计算所述第二特征集合中的所有余下的局部特征分别到这n个种子点的距离(例如欧氏距离、余弦距离),将所有余下的局部特征分别划分到其距离最近的种子点的点群(例如局部特征Pi离种子点Si最近,那么Pi被划分到Si点群),得到n个点群;
c、计算得到的每个点群的中心(即中心点特征向量),将种子点移动到点群的中心;
d、重复步骤b和c,直到种子点不再移动后,此时的n个点群则为分类结果。其中,点群中的每一个点均代表一张样本图片的局部特征。
此外,本申请还提出一种混乱样本整理系统。
请参阅图3,是本申请混乱样本整理系统10较佳实施例的运行环境示意图。
在本实施例中,混乱样本整理系统10安装并运行于电子装置1中。电子装置1可以是桌上型计算机、笔记本、掌上电脑及服务器等计算设备。该电子装置1可包括,但不仅限于,存储器11、处理器12及显示器13。图3仅示出了具有组件11-13的电子装置1,但是应理解的是,并不要求实施所有示出的组件,可以替代的实施更多或者更少的组件。
存储器11在一些实施例中可以是电子装置1的内部存储单元,例如该电子装置1的硬盘或内存。存储器11在另一些实施例中也可以是电子装置1的外部存储设备,例如电子装置1上配备的插接式硬 盘,智能存储卡(Smart Media Card,SMC),安全数字(Secure Digital,SD)卡,闪存卡(Flash Card)等。进一步地,存储器11还可以既包括电子装置1的内部存储单元也包括外部存储设备。存储器11用于存储安装于电子装置1的应用软件及各类数据,例如混乱样本整理系统10的程序代码等。存储器11还可以用于暂时地存储已经输出或者将要输出的数据。
处理器12在一些实施例中可以是一中央处理器(Central Processing Unit,CPU),微处理器或其他数据处理芯片,用于运行存储器11中存储的程序代码或处理数据,例如执行混乱样本整理系统10等。
显示器13在一些实施例中可以是LED显示器、液晶显示器、触控式液晶显示器以及OLED(Organic Light-Emitting Diode,有机发光二极管)触摸器等。显示器13用于显示在电子装置1中处理的信息以及用于显示可视化的用户界面,例如业务定制界面等。电子装置1的部件11-13通过系统总线相互通信。
请参阅图4,是本申请混乱样本整理系统10一实施例的程序模块图。在本实施例中,混乱样本整理系统10可以被分割成一个或多个模块,一个或者多个模块被存储于存储器11中,并由一个或多个处理器(本实施例为处理器12)所执行,以完成本申请。例如,在图4中,混乱样本整理系统10可以被分割成获取模块101、提取模块102、降维模块103及第一聚类模块104。本申请所称的模块是指能够完成特定功能的一系列计算机程序指令段,比程序更适合于描述混乱样本整理系统10在电子装置1中的执行过程,其中:
获取模块101,用于获取待整理的样本图片,将获取的样本图片统一调整至预设尺寸;
首先,系统从样本库中获取待整理的样本图片,获取到的样本图片的尺寸大小可能不一致,因此先对获取的样本图片进行尺寸调整,将所有样本图片的尺寸统一调整至预设尺寸,为后续计算作准备;所述预设尺寸应该小于所有样本图片中的最小样本图片的尺寸。例如,获取的样本图片中存在800像素×600像素、480像素×800像素等 各种尺寸的样本图片,统一将这些样本图片的尺寸调整为256像素×256像素(即预设尺寸)。
提取模块102,用于对调整尺寸后的所有样本图片进行局部特征提取,以得到第一特征集合;
系统在调整好样本图片的尺寸后,对每一张样本图片分别进行局部特征提取,提取出所有样本图片的局部特征形成第一特征集合。本实施例中,局部特征可以为LBP(Local binary patterns)特征、HOG(Histogram of Oriented Gradient)特征、SURF(Speeded Up Robust Features)特征,等等。优选地,本实施例中,对样本图片的局部特征的提取方式可以为:1、针对每一张样本图片,提取一个局部特征;2、针对每一张样本图片,先将该样本图片划分为预设的m(例如m=4)个图像区域,然后从各个图像区域中分别提取一个局部特征,再计算出从这m个图像区域提取出的m个局部特征的平均局部特征,将该平均局部特征作为该样本图片的局部特征,这样该样本图片的局部特征可以更好的代表该样本图片。
降维模块103,用于对所述第一特征集合中的局部特征进行PCA降维处理,得到降维处理后的第二特征集合;
对第一特征集合中的所有局部特征均通过PCA(Principal Component Analysis,主成分分析)算法进行降维处理,降维后的所有局部特征形成第二特征集合。从样本图片提取的局部特征是由一系列特征组成的高维向量(例如,维度达到5000),其中,某些维度的元素在所有样本图片中都相等或者彼此差距不大,那么这些维度元素对于区分样本图片的贡献度就很小;通过PCA降维的目的就是为去掉第一特征集合中这些对区分样本图片贡献度很小的维度元素,并找到第一特征集合中那些方差大的维度元素(对区分样本图片贡献度大的维度元素)。
第一聚类模块104,用于采用预设类型的聚类算法将第二特征集合中的局部特征聚类成预先设置的n类,根据第二特征集合中的局部特征聚类结果,将获取到的所有样本图片分成n类。
采用预设类型的聚类算法对第二特征集合中的局部特征进行聚 类,以将第二特征集合中的局部特征聚类成预先设置的n(例如n=3)类;然后根据第二特征集合中的局部特征的聚类结果,将各个局部特征所属的样本图片对应的分成n类,如此则完成对待处理的样本图片的分类整理。
本实施例技术方案,首先对所有样本图片进行尺寸统一规范调整,然后提取调整尺寸后的每张样本图片的局部特征,组成第一特征集合,再将第一特征结合中的局部特征进行PCA降维处理,得到降维处理后的局部特征形成的第二特征集合,最终通过预设的聚类算法对第二特征集合中的局部特征进行聚类,将局部特征聚类成预设的n类,根据局部特征的分类,对应的将样本图片进行分类整理为n类,即完成样本图片的整理分类。与现有技术相比,本方案通过系统对样本图片进行自动的分类整理,样本整理速度快、效率高,解决了人工整理样本耗时太长而影响样本训练工作的进度的问题,并且降低了人工成本。
参阅图5,在本实施例中,所述混乱样本整理系统还包括:
样本分配模块105,用于计算获取到的样本图片数量,根据预设的样本数量区间与文件夹数量的映射关系,确定当前样本图片数量对应的文件夹数量,将调整尺寸后的所有样本图片均匀分配至确定数量的文件夹中;
当样本图片数量巨大(例如样本图片多至几万或几十万张)时,为了减少计算量,本实施例将样本图片划分至多个文件夹,以文件夹为单位进行后续的计算、处理。系统预先设置了样本数量区间与分配文件夹数量的映射关系表,系统在计算出获取的样本图片数量后,根据该样本图片数量所属的样本数量区间,通过查询映射关系表就可确定该样本图片数量对应的需分配的文件夹数量,在确定文件夹数量后,将经过尺寸统一调整后的所有样本图片以均匀的方式分配至确定数量的文件夹中。比如有10万张样本图片,9至11万的样本数量区间对应的文件夹数量为10个,则将这10万张样本图片分开放至folder1、folder2……folder10这10个文件夹中,每个文件夹中放1万 张。需要说明的是,本实施例仅仅是优选均匀分配方式为例,在其它实施例中,还可以采取其它方式将样本图片分配至各个文件夹中。另外,当样本图片数量不是很巨大(例如样本图片仅有几千张)时,可以不对样本图片进行文件夹的划分,或该样本图片数量对应为一个文件夹,即无需对样本图片进行分配。
将第一聚类模块104替换为第二聚类模块106,所述第二聚类模块106用于采用预设类型的聚类算法将第二特征集合中的局部特征聚类成预先设置的n类,根据聚类后的结果将每个文件夹中的样本图片分成n类。
优选地,本实施例中第一聚类模块104和第二聚类模块106采用的预设类型的聚类算法为k-means算法,所述将第二特征集合中的局部特征聚类成预设的n个类别的步骤包括:
a、随机在所述第二特征集合中选取n(例如n=3)个局部特征作为种子点;
b、计算所述第二特征集合中的所有余下的局部特征分别到这n个种子点的距离(例如欧氏距离、余弦距离),将所有余下的局部特征分别划分到其距离最近的种子点的点群(例如局部特征Pi离种子点Si最近,那么Pi被划分到Si点群),得到n个点群;
c、计算得到的每个点群的中心(即中心点特征向量),将种子点移动到点群的中心;
d、重复步骤b和c,直到种子点不再移动后,此时的n个点群则为分类结果。其中,点群中的每一个点均代表一张样本图片的局部特征。
进一步地,本申请还提出一种计算机可读存储介质,所述计算机可读存储介质存储有混乱样本整理系统,所述混乱样本整理系统可被至少一个处理器执行,以使所述至少一个处理器执行上述任一实施例中的混乱样本整理方法。
以上所述仅为本申请的优选实施例,并非因此限制本申请的专利范围,凡是在本申请的申请构思下,利用本申请说明书及附图内容所作的等效结构变换,或直接/间接运用在其他相关的技术领域均包括在本申请的专利保护范围内。
Claims (20)
- 一种电子装置,其特征在于,所述电子装置包括存储器和处理器,所述存储器上存储有可在所述处理器上运行的混乱样本整理系统,所述混乱样本整理系统被所述处理器执行时实现如下步骤:A1、获取待整理的样本图片,将获取的样本图片统一调整至预设尺寸;B1、对调整尺寸后的所有样本图片进行局部特征提取,以得到第一特征集合;C1、对所述第一特征集合中的局部特征进行PCA降维处理,得到降维处理后的第二特征集合;D1、采用预设类型的聚类算法将第二特征集合中的局部特征聚类成预先设置的n类,根据第二特征集合中的局部特征聚类结果,将获取到的所有样本图片分成n类。
- 如权利要求1所述的电子装置,其特征在于,所述将第二特征集合中的局部特征聚类成预设的n个类别的步骤包括:a、随机在所述第二特征集合中选取n个局部特征作为种子点;b、计算所述第二特征集合中的所有余下的局部特征分别到这n个种子点的距离,将所有余下的局部特征分别划分到其距离最近的种子点的点群,得到n个点群;c、计算得到的每个点群的中心,将种子点移动到点群的中心;d、重复步骤b和c,直到种子点不再移动后,此时的n个点群则为分类结果。
- 如权利要求1所述的电子装置,其特征在于,在所述步骤B中,样本图片的局部特征的提取方式为:针对每一张样本图片,提取一个局部特征;或者,针对每一张样本图片,将该样本图片划分为预设的m个图像区域,从各个图像区域中分别提取一个局部特征,再计算m个图像区域的平均局部特征,该平均局部特征作为该样本图片的局部特征。
- 如权利要求3所述的电子装置,其特征在于,所述将第二特征集合中的局部特征聚类成预设的n个类别的步骤包括:a、随机在所述第二特征集合中选取n个局部特征作为种子点;b、计算所述第二特征集合中的所有余下的局部特征分别到这n个种子点的距离,将所有余下的局部特征分别划分到其距离最近的种子点的点群,得到n个点群;c、计算得到的每个点群的中心,将种子点移动到点群的中心;d、重复步骤b和c,直到种子点不再移动后,此时的n个点群则为分类结果。
- 如权利要求1所述的电子装置,其特征在于,于所述步骤B1之前,所述处理器还用于执行所述混乱样本整理系统,以实现步骤:计算获取到的样本图片数量,根据预设的样本数量区间与文件夹数量的映射关系,确定当前样本图片数量对应的文件夹数量,将调整尺寸后的所有样本图片均匀分配至确定数量的文件夹中;所述步骤D1替换为:采用预设类型的聚类算法将第二特征集合中的局部特征聚类成预先设置的n类,根据第二特征集合中的局部特征聚类结果,将每个文件夹中的样本图片分成n类。
- 如权利要求5所述的电子装置,其特征在于,所述将第二特征集合中的局部特征聚类成预设的n个类别的步骤包括:a、随机在所述第二特征集合中选取n个局部特征作为种子点;b、计算所述第二特征集合中的所有余下的局部特征分别到这n个种子点的距离,将所有余下的局部特征分别划分到其距离最近的种子点的点群,得到n个点群;c、计算得到的每个点群的中心,将种子点移动到点群的中心;d、重复步骤b和c,直到种子点不再移动后,此时的n个点群则为分类结果。
- 如权利要求5所述的电子装置,其特征在于,在所述步骤B中,样本图片的局部特征的提取方式为:针对每一张样本图片,提取一个局部特征;或者,针对每一张样本图片,将该样本图片划分为预设的m个图像区域,从各个图像区域中分别提取一个局部特征,再计算m个图像区域的平均局部特征,该平均局部特征作为该样本图片的局部特征。
- 如权利要求7所述的电子装置,其特征在于,所述将第二特征集合中的局部特征聚类成预设的n个类别的步骤包括:a、随机在所述第二特征集合中选取n个局部特征作为种子点;b、计算所述第二特征集合中的所有余下的局部特征分别到这n个种子点的距离,将所有余下的局部特征分别划分到其距离最近的种子点的点群,得到n个点群;c、计算得到的每个点群的中心,将种子点移动到点群的中心;d、重复步骤b和c,直到种子点不再移动后,此时的n个点群则为分类结果。
- 一种混乱样本整理方法,其特征在于,该方法包括步骤:A2、获取待整理的样本图片,将获取的样本图片统一调整至预设尺寸;B2、对调整尺寸后的所有样本图片进行局部特征提取,以得到第一特征集合;C2、对所述第一特征集合中的局部特征进行PCA降维处理,得到第二特征集合;D2、采用预设类型的聚类算法将第二特征集合中的局部特征聚类成预设的n个类别,根据第二特征集合中的局部特征聚类结果,将获取到的所有样本图片分成n类。
- 如权利要求9所述的混乱样本整理方法,其特征在于,所述将第二特征集合中的局部特征聚类成预设的n个类别的步骤包括:a、随机在所述第二特征集合中选取n个局部特征作为种子点;b、计算所述第二特征集合中的所有余下的局部特征分别到这n个种子点的距离,将所有余下的局部特征分别划分到其距离最近的种子点的点群,得到n个点群;c、计算得到的每个点群的中心,将种子点移动到点群的中心;d、重复步骤b和c,直到种子点不再移动后,此时的n个点群则为分类结果。
- 如权利要求9所述的混乱样本整理方法,其特征在于,在所述步骤B中,样本图片的局部特征的提取方式为:针对每一张样本图片,提取一个局部特征;或者,针对每一张样本图片,将该样本图片划分为预设的m个图像区域,从各个图像区域中分别提取一个局部特征,再计算m个图像区域的平均局部特征,该平均局部特征作为该样本图片的局部特征。
- 如权利要求11所述的混乱样本整理方法,其特征在于,所述将第二特征集合中的局部特征聚类成预设的n个类别的步骤包括:a、随机在所述第二特征集合中选取n个局部特征作为种子点;b、计算所述第二特征集合中的所有余下的局部特征分别到这n个种子点的距离,将所有余下的局部特征分别划分到其距离最近的种子点的点群,得到n个点群;c、计算得到的每个点群的中心,将种子点移动到点群的中心;d、重复步骤b和c,直到种子点不再移动后,此时的n个点群则为分类结果。
- 如权利要求9所述的混乱样本整理方法,其特征在于,于所述步骤B2之前,所述混乱样本整理方法还包括:计算获取到的样本图片数量,根据预设的样本数量区间与文件夹数量的映射关系,确定当前样本图片数量对应的文件夹数量,将调整尺寸后的所有样本图片均匀分配至确定数量的文件夹中;所述步骤D2替换为:采用预设类型的聚类算法将第二特征集合中的局部特征聚类成预先设置的n类,根据第二特征集合中的局部特征聚类结果,将每个文件夹中的样本图片分成n类。
- 如权利要求13所述的混乱样本整理方法,其特征在于,所述将第二特征集合中的局部特征聚类成预设的n个类别的步骤包括:a、随机在所述第二特征集合中选取n个局部特征作为种子点;b、计算所述第二特征集合中的所有余下的局部特征分别到这n个种子点的距离,将所有余下的局部特征分别划分到其距离最近的种子点的点群,得到n个点群;c、计算得到的每个点群的中心,将种子点移动到点群的中心;d、重复步骤b和c,直到种子点不再移动后,此时的n个点群则为分类结果。
- 如权利要求13所述的混乱样本整理方法,其特征在于,在所述步骤B中,样本图片的局部特征的提取方式为:针对每一张样本图片,提取一个局部特征;或者,针对每一张样本图片,将该样本图片划分为预设的m个图像区域,从各个图像区域中分别提取一个局部特征,再计算m个图像区域的平均局部特征,该平均局部特征作为该样本图片的局部特征。
- 如权利要求15所述的混乱样本整理方法,其特征在于,所述将第二特征集合中的局部特征聚类成预设的n个类别的步骤包括:a、随机在所述第二特征集合中选取n个局部特征作为种子点;b、计算所述第二特征集合中的所有余下的局部特征分别到这n个种子点的距离,将所有余下的局部特征分别划分到其距离最近的种子点的点群,得到n个点群;c、计算得到的每个点群的中心,将种子点移动到点群的中心;d、重复步骤b和c,直到种子点不再移动后,此时的n个点群则为分类结果。
- 一种计算机可读存储介质,其特征在于,所述计算机可读存储介质存储有混乱样本整理系统,所述混乱样本整理方法系统可被至少一个处理器执行,以使所述至少一个处理器执行如下步骤:A3、获取待整理的样本图片,将获取的样本图片统一调整至预设尺寸;B3、对调整尺寸后的所有样本图片进行局部特征提取,以得到第一特征集合;C3、对所述第一特征集合中的局部特征进行PCA降维处理,得 到第二特征集合;D3、采用预设类型的聚类算法将第二特征集合中的局部特征聚类成预设的n个类别,根据第二特征集合中的局部特征聚类结果,将获取到的所有样本图片分成n类。
- 如权利要求17所述的计算机可读存储介质,其特征在于,所述样本图片的局部特征的提取方式为:针对每一张样本图片,提取一个局部特征;或者,针对每一张样本图片,将该样本图片划分为预设的m个图像区域,从各个图像区域中分别提取一个局部特征,再计算m个图像区域的平均局部特征,该平均局部特征作为该样本图片的局部特征。
- 如权利要求17所述的计算机可读存储介质,其特征在于,于所述步骤B3之前,所述处理器还用于执行所述混乱样本整理系统,以实现步骤:计算获取到的样本图片数量,根据预设的样本数量区间与文件夹数量的映射关系,确定当前样本图片数量对应的文件夹数量,将调整尺寸后的所有样本图片均匀分配至确定数量的文件夹中;所述步骤D3替换为:采用预设类型的聚类算法将第二特征集合中的局部特征聚类成预先设置的n类,根据第二特征集合中的局部特征聚类结果,将每个文件夹中的样本图片分成n类。
- 如权利要求17所述的计算机可读存储介质,其特征在于,所述将第二特征集合中的局部特征聚类成预设的n个类别的步骤包括:a、随机在所述第二特征集合中选取n个局部特征作为种子点;b、计算所述第二特征集合中的所有余下的局部特征分别到这n个种子点的距离,将所有余下的局部特征分别划分到其距离最近的种子点的点群,得到n个点群;c、计算得到的每个点群的中心,将种子点移动到点群的中心;d、重复步骤b和c,直到种子点不再移动后,此时的n个点群 则为分类结果。
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711025204.0 | 2017-10-27 | ||
| CN201711025204.0A CN108427955B (zh) | 2017-10-27 | 2017-10-27 | 电子装置、混乱样本整理方法和计算机可读存储介质 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| WO2019080430A1 true WO2019080430A1 (zh) | 2019-05-02 |
Family
ID=63155684
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| PCT/CN2018/077671 Ceased WO2019080430A1 (zh) | 2017-10-27 | 2018-02-28 | 电子装置、混乱样本整理方法和计算机可读存储介质 |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN108427955B (zh) |
| WO (1) | WO2019080430A1 (zh) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101859320A (zh) * | 2010-05-13 | 2010-10-13 | 复旦大学 | 一种基于多特征签名的海量图像检索方法 |
| CN105005789A (zh) * | 2015-07-01 | 2015-10-28 | 北京理工大学 | 一种基于视觉词汇的遥感图像地物分类方法 |
| CN105335702A (zh) * | 2015-10-15 | 2016-02-17 | 浙江捷尚视觉科技股份有限公司 | 一种基于统计学习的卡口车型识别方法 |
| CN106250821A (zh) * | 2016-07-20 | 2016-12-21 | 南京邮电大学 | 一种聚类再分类的人脸识别方法 |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP5526955B2 (ja) * | 2010-04-09 | 2014-06-18 | ソニー株式会社 | 顔クラスタリング装置、顔クラスタリング方法、及びプログラム |
| CN105389593B (zh) * | 2015-11-16 | 2019-01-11 | 上海交通大学 | 基于surf特征的图像物体识别方法 |
-
2017
- 2017-10-27 CN CN201711025204.0A patent/CN108427955B/zh active Active
-
2018
- 2018-02-28 WO PCT/CN2018/077671 patent/WO2019080430A1/zh not_active Ceased
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101859320A (zh) * | 2010-05-13 | 2010-10-13 | 复旦大学 | 一种基于多特征签名的海量图像检索方法 |
| CN105005789A (zh) * | 2015-07-01 | 2015-10-28 | 北京理工大学 | 一种基于视觉词汇的遥感图像地物分类方法 |
| CN105335702A (zh) * | 2015-10-15 | 2016-02-17 | 浙江捷尚视觉科技股份有限公司 | 一种基于统计学习的卡口车型识别方法 |
| CN106250821A (zh) * | 2016-07-20 | 2016-12-21 | 南京邮电大学 | 一种聚类再分类的人脸识别方法 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108427955B (zh) | 2022-02-01 |
| CN108427955A (zh) | 2018-08-21 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| EP2805262B1 (en) | Image index generation based on similarities of image features | |
| US8392484B2 (en) | Shape based picture search | |
| CN110431560B (zh) | 目标人物的搜索方法和装置、设备和介质 | |
| US10373014B2 (en) | Object detection method and image search system | |
| WO2019080411A1 (zh) | 电子装置、人脸图像聚类搜索方法和计算机可读存储介质 | |
| CN102521366A (zh) | 融合分类与全局索引的图像检索方法和图像检索系统 | |
| CN111325245A (zh) | 重复图像识别方法、装置、电子设备及计算机可读存储介质 | |
| WO2015035477A1 (en) | Image searching method and apparatus | |
| US9501554B2 (en) | Image processing system, image processing method, and image processing program | |
| CN104282008B (zh) | 对图像进行纹理分割的方法和装置 | |
| CN110347855A (zh) | 画作推荐方法、终端设备、服务器、计算机设备及介质 | |
| CN113780116B (zh) | 发票分类方法、装置、计算机设备和存储介质 | |
| CN111353062A (zh) | 一种图像检索方法、装置以及设备 | |
| CN114881698A (zh) | 广告合规审核方法、装置、电子设备及存储介质 | |
| CN111931229B (zh) | 一种数据识别方法、装置和存储介质 | |
| CN110717806A (zh) | 产品信息推送方法、装置、设备及存储介质 | |
| CN106650804A (zh) | 一种基于深度学习特征的人脸样本清洗方法及系统 | |
| Sunitha et al. | Novel content based medical image retrieval based on BoVW classification method | |
| CN104966109B (zh) | 医疗化验单图像分类方法及装置 | |
| CN112668583A (zh) | 图像识别方法、装置以及电子设备 | |
| CN107688744A (zh) | 基于图像特征匹配的恶意文件分类方法及装置 | |
| US9715624B1 (en) | Document image segmentation based on pixel classification | |
| WO2019080430A1 (zh) | 电子装置、混乱样本整理方法和计算机可读存储介质 | |
| CN111695441A (zh) | 图像文档处理方法、装置及计算机可读存储介质 | |
| JP7507331B1 (ja) | レイアウト解析システム、レイアウト解析方法、及びプログラム |
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: 18870704 Country of ref document: EP Kind code of ref document: A1 |
|
| NENP | Non-entry into the national phase |
Ref country code: DE |
|
| 32PN | Ep: public notification in the ep bulletin as address of the adressee cannot be established |
Free format text: NOTING OF LOSS OF RIGHTS (EPO FORM 1205A DATED 25.09.2020) |
|
| 122 | Ep: pct application non-entry in european phase |
Ref document number: 18870704 Country of ref document: EP Kind code of ref document: A1 |