WO2017143979A1 - 图像的检索方法及装置 - Google Patents

图像的检索方法及装置 Download PDF

Info

Publication number
WO2017143979A1
WO2017143979A1 PCT/CN2017/074356 CN2017074356W WO2017143979A1 WO 2017143979 A1 WO2017143979 A1 WO 2017143979A1 CN 2017074356 W CN2017074356 W CN 2017074356W WO 2017143979 A1 WO2017143979 A1 WO 2017143979A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
histogram
feature
images
retrieved
Prior art date
Application number
PCT/CN2017/074356
Other languages
English (en)
French (fr)
Inventor
朱海涛
Original Assignee
中兴通讯股份有限公司
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 中兴通讯股份有限公司 filed Critical 中兴通讯股份有限公司
Publication of WO2017143979A1 publication Critical patent/WO2017143979A1/zh

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00Image analysis
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/50Information retrieval; Database structures therefor; File system structures therefor of still image data
    • G06F16/58Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
    • G06F16/583Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content

Definitions

  • the present disclosure relates to the field of image retrieval, for example, to a method and apparatus for searching an image.
  • Image retrieval techniques in the related art include text-based image retrieval technology and image-based image retrieval technology.
  • the text-based image retrieval technology uses text retrieval technology, artificially adds annotations to each image in the image library, which is used to describe the information of the image, and the user also retrieves a kind of image through text when searching, such as Baidu pictures can support this.
  • An image retrieval method With the development of technologies such as digital image processing, pattern recognition, and machine learning, image-based image retrieval technology emerged.
  • Image-based image retrieval technology applies a large number of principles and knowledge in the fields of digital image processing, pattern recognition and machine learning. The image is extracted by a specific algorithm, and the similarity between images is calculated by the extracted features, and returned according to the similarity. A similar image in the image library completes the entire image retrieval process.
  • Text-based image retrieval Although the subject of the searched images mostly conforms to the semantics of the image to be retrieved input by the user, each image in the image library of the text-based image retrieval method needs to be manually labeled, which consumes a large amount of manpower. This approach is becoming less and less practical in the context of faster and faster Internet image updates and higher labor costs.
  • the image-based image retrieval system in the related art mostly searches for similar images by calculating the similarity of the image on one or several features, such as color features, histogram features, gradient features, geometric features, etc., but these features Most of them are not semantic, resulting in a low degree of matching between the retrieved image and the input image.
  • the present disclosure provides an image retrieval method and apparatus, which avoids the phenomenon that the image retrieval method is single and the matching degree is not high in the related art.
  • the present disclosure provides an image retrieval method, including: extracting a plurality of feature descriptors for characterizing an image feature attribute on an image to be retrieved; mapping the extracted feature descriptor to a pre-generated vocabulary, and counting a histogram on the vocabulary; and calculating a similarity between the histogram of the image to be retrieved and the histogram of each image in the pre-existing image library, and retrieving the image to be retrieved from the image library
  • the histogram's similarity is greater than all images of the preset threshold.
  • the method before extracting the plurality of feature descriptors for characterizing the image feature attributes on the image to be retrieved, the method further includes: collecting multiple types of images, and preprocessing the plurality of types of images Normalized plurality of types of images; and extracting feature descriptors of the normalized plurality of types of images, and performing clustering processing on the feature descriptors of the plurality of types of images for generating A vocabulary that maps feature descriptors for a variety of different image feature attributes.
  • the method further includes: extracting feature descriptions of the normalized plurality of types of images Mapping all feature descriptors on each of the plurality of types of images onto a vocabulary, and counting histograms of each of the words on the vocabulary; and placing the histograms The graph is normalized to obtain a normalized histogram of each of the images on the vocabulary.
  • the calculating the similarity between the histogram of the image to be retrieved and the histogram of each image in the pre-existing image library comprises: counting each item in the histogram of the image to be retrieved to represent different image feature attributes Calculating a ratio of the number of feature descriptors to the sum of all feature descriptors in the histogram of the image to be retrieved; and calculating a histogram of the image to be retrieved and each image in the pre-existing image library according to the scale value The similarity of the histograms.
  • the sum of the scale values of the feature descriptors of all the different image feature attributes in each of the histograms is one.
  • the present disclosure also provides an image retrieval device, comprising: a feature extraction module, configured to extract a plurality of feature descriptors for characterizing image feature attributes on the image to be retrieved; and a first mapping module configured to be extracted The feature descriptor is mapped onto the pre-generated vocabulary and the histogram on the vocabulary is counted; and the similarity calculation module is configured to calculate a histogram of the image to be retrieved and each image in the pre-existing image library The similarity of the histograms, and all images whose similarities with the histograms of the images to be retrieved are greater than a preset threshold are retrieved from the image library.
  • the apparatus further includes: a first processing module, configured to collect multiple types of images before extracting a plurality of feature descriptors for characterizing image feature attributes on the image to be retrieved, and The plurality of types of images are pre-processed to obtain the normalized plurality of types of images; and the second processing module is configured to extract feature descriptors of the normalized plurality of types of images, and to The feature descriptors of the various types of images perform clustering processing to generate a vocabulary for mapping feature descriptors of a plurality of different image feature attributes.
  • a first processing module configured to collect multiple types of images before extracting a plurality of feature descriptors for characterizing image feature attributes on the image to be retrieved, and The plurality of types of images are pre-processed to obtain the normalized plurality of types of images
  • the second processing module is configured to extract feature descriptors of the normalized plurality of types of images, and to The feature descriptors of the various types of images perform clustering processing to generate a vocabulary for mapping
  • the apparatus further includes: an extracting module, configured to extract the normalized multiple types after performing clustering processing on the feature descriptors of the plurality of types of images to generate the vocabulary a feature descriptor of all images in the image library; a second mapping module configured to map all feature descriptors on each of the plurality of types of images onto the vocabulary, and to count each of the A histogram of each vocabulary of the image on the vocabulary; and a normalization module configured to normalize the histogram to obtain a normalized histogram of each of the images on the vocabulary.
  • an extracting module configured to extract the normalized multiple types after performing clustering processing on the feature descriptors of the plurality of types of images to generate the vocabulary a feature descriptor of all images in the image library
  • a second mapping module configured to map all feature descriptors on each of the plurality of types of images onto the vocabulary, and to count each of the A histogram of each vocabulary of the image on the vocabulary
  • a normalization module configured to normalize the histogram to obtain
  • the similarity calculation module includes: a statistical unit, configured to calculate a histogram of the number of feature descriptors of each item representing different image feature attributes in the histogram of the image to be retrieved, and the image to be retrieved And a similarity calculation unit configured to calculate a similarity between the histogram of the image to be retrieved and the histogram of each image in the pre-existing image library according to the scale value.
  • the sum of the scale values of the feature descriptors of all the different image feature attributes in each of the histograms is one.
  • the present disclosure also provides a non-transitory computer readable storage medium storing computer executable instructions arranged to perform the above method.
  • the present disclosure also provides an electronic device, including:
  • At least one processor At least one processor
  • the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to cause the at least one processor to perform the method described above.
  • the present disclosure by extracting a plurality of feature descriptors for characterizing an image feature attribute on an image to be retrieved, mapping the extracted feature descriptor to a pre-generated vocabulary, and counting the image to be retrieved in the vocabulary
  • the upper histogram by calculating the similarity between the histogram of the image to be retrieved and the histogram of each image in the pre-existing image library, and retrieving from the image library the similarity with the histogram of the image to be retrieved is greater than the pre- Set all the images of the threshold to avoid the single retrieval method of the image in the related art.
  • the phenomenon of low matching by extracting a plurality of feature descriptors for characterizing an image feature attribute on an image to be retrieved, mapping the extracted feature descriptor to a pre-generated vocabulary, and counting the image to be retrieved in the vocabulary
  • the upper histogram by calculating the similarity between the histogram of the image to be retrieved and the histogram of each image in the pre-existing image library, and retriev
  • FIG. 1 is a flowchart of a method of retrieving an image of an embodiment of the present disclosure
  • FIG. 2 is a block diagram showing the structure of an image retrieval apparatus according to an alternative embodiment of the present disclosure
  • FIG. 3 is a block diagram 1 of an optional structure of an image retrieval device according to an alternative embodiment of the present disclosure
  • FIG. 4 is a block diagram 2 of an optional structure of an image retrieval device according to an alternative embodiment of the present disclosure
  • FIG. 5 is a flow chart of a vocabulary training method of an alternative embodiment of the present disclosure.
  • FIG. 6 is a flowchart of a method for establishing an image library according to an alternative embodiment of the present disclosure
  • FIG. 7 is a flowchart of an image retrieval method of an alternative embodiment of the present disclosure.
  • FIG. 8 is a schematic structural diagram of hardware of an electronic device according to an embodiment of the present disclosure.
  • FIG. 1 is a flowchart of an image retrieval method according to an embodiment of the present disclosure.
  • step 110 a plurality of feature descriptors for characterizing image feature attributes on the image to be retrieved are extracted.
  • step 120 the extracted feature descriptors are mapped onto a pre-generated vocabulary, and a histogram of the image to be retrieved on the vocabulary is counted.
  • step 130 calculating a similarity between the histogram of the image to be retrieved and the histogram of each image in the pre-existing image library, and retrieving from the image library the similarity with the histogram of the image to be retrieved is greater than All images of the preset threshold.
  • the image whose histogram similarity is greater than the preset threshold that is, by mapping the feature descriptor of the image into the vocabulary and representing different types of image feature attributes by the histogram, by comparing the histogram of the image to be retrieved
  • the image with the similarity of the histogram of the image to be retrieved is higher than the preset threshold, and the phenomenon that the image retrieval method is single and the matching degree is not high in the related art is avoided.
  • the method in this embodiment may further include:
  • a feature descriptor of the normalized plurality of types of images is extracted, and feature descriptors of the plurality of types of images are clustered to generate a vocabulary for mapping feature descriptors of the plurality of different image feature attributes.
  • the image may include two types of images of a person and a scene; preprocessing the collected image to normalize the image size, and extracting the characteristics of the dense distribution of important information
  • the important information of the feature descriptor of the feature point can represent the importance degree of the feature descriptor. The smaller the distance from the position of the feature point to the center of the image, the more important the feature descriptor of the feature point.
  • the method in this embodiment may further include:
  • the histogram is normalized to obtain a normalized histogram of each image on the vocabulary.
  • preprocessing the image in the image library to normalize the image size For example, preprocessing the image in the image library to normalize the image size; extracting the feature descriptor of the corresponding point on each image by using the dense distribution of the important information; mapping the extracted feature descriptor to the pre-generated a histogram on the vocabulary, and a histogram of the image to be retrieved on the vocabulary; a sum of the number of occurrences of all feature descriptors in the statistical histogram, and each item in the histogram represents a feature of a different image feature attribute
  • the number of occurrences of the descriptor is divided by the sum, and the scale value is obtained, ensuring that the sum of the scale values of the feature descriptors representing the different feature attributes of the processed histogram is 1, and the processed histogram data is saved to a file or In the database.
  • calculating the similarity between the histogram of the image to be retrieved and the histogram of all the images in the pre-existing image library includes:
  • a software product stored in a storage medium (such as a read-only memory (ROM), a random storage memory (Random-Access Memory, RAM). , a disk, an optical disk, including one or more instructions for causing a terminal device (which may be a mobile phone, a computer, a server, or a network device, etc.) to perform the method of the embodiments of the present disclosure.
  • an image retrieval device is further provided, which can implement the above embodiments and alternative embodiments.
  • the term "module” may implement a combination of at least one of software and hardware for a predetermined function.
  • the apparatus described in the following embodiments may be implemented in software, it may be implemented in hardware, or a combination of software and hardware.
  • the apparatus includes a feature extraction module 22, a first mapping module 24, and a similarity calculation module 26.
  • Feature extraction module 22 is configured to extract a plurality of features on the image to be retrieved for characterizing image feature attributes Sign the descriptor.
  • the first mapping module 24 is coupled to the feature extraction module 22, and is configured to map the extracted feature descriptors to the pre-generated vocabulary and count the histogram of the image to be retrieved on the vocabulary.
  • the similarity calculation module 26 is coupled to the first mapping module 24 and configured to calculate a similarity between the histogram of the image to be retrieved and the histogram of each image in the pre-existing image library, and retrieve and describe the image from the image library.
  • the images of the histogram to be retrieved have similarities that are greater than the preset thresholds.
  • FIG. 3 is a block diagram of an optional structure of an image retrieval apparatus according to an alternative embodiment of the present disclosure.
  • the apparatus may further include: a first processing module 32 and a second processing module 34.
  • the first processing module 32 is configured to collect multiple types of images before extracting a plurality of feature descriptors for characterizing the image feature attributes on the image to be retrieved, and preprocess the plurality of types of images to obtain a normalized Multiple types of images.
  • the second processing module 34 is coupled to the first processing module 32, and is configured to extract feature descriptors of the normalized plurality of types of images, and perform clustering processing on feature descriptors of the plurality of types of images.
  • FIG. 4 is a block diagram of an optional structure of an image retrieval apparatus according to an alternative embodiment of the present disclosure.
  • the apparatus may further include: an extraction module 42, a second mapping module 44, and a normalization module 46.
  • the extraction module 42 is configured to extract the feature descriptors of the normalized plurality of types of images after performing clustering processing on the feature descriptors of the plurality of types of images to generate the vocabulary.
  • the second mapping module 44 is coupled to the extraction module 42 and configured to map all the feature descriptors on each of the plurality of types of images to the vocabulary, and count each image on the vocabulary A histogram of each vocabulary.
  • the normalization module 46 is coupled to the second mapping module 44 and is configured to normalize the histogram to obtain a normalized histogram of each image on the vocabulary.
  • the similarity calculation module 26 in FIG. 2 in this embodiment may further include: a statistics unit and a similarity calculation unit.
  • the statistical unit is configured to count the number of feature descriptors of each item in the histogram of the image to be retrieved representing different image feature attributes as a proportion of the sum of all feature descriptors in the histogram of the image to be retrieved.
  • the similarity calculation unit is configured to calculate a similarity between the histogram of the image to be retrieved and the histogram of each image in the pre-existing image library according to the scale value.
  • the sum of the scale values of the feature descriptors of all the different image feature attributes in the histogram in the present embodiment is 1.
  • the above modules can be implemented by software or hardware. When implemented by hardware, the modules are all located in the same processor, or the modules are respectively located in multiple processors.
  • the present disclosure proposes a word bag model image retrieval method based on the dense distribution characteristics of important information.
  • the optional embodiment generates a vocabulary by extracting a dense feature of the image and generating a certain amount of representative feature descriptors by clustering processing.
  • image retrieval is accomplished by measuring the similarity of the histogram of the picture to be retrieved to the histogram on each vocabulary in the image library.
  • the word bag model involved in this alternative embodiment can extract certain semantic information, and can extract more important information by using the dense distribution feature of the important information on the image, ignoring certain secondary information.
  • the dense feature may refer to a dense distribution feature of the important information on the image (ie, there are more important information near the center of the image, and some feature points may be selected near the center of the image), and feature points are extracted, and the selected features are extracted.
  • the feature descriptor of the point All feature points and feature descriptors extracted by the above method are called dense features of the image.
  • the retrieval method of the alternative embodiment may include training a vocabulary, constructing a library of images to be retrieved, and image retrieval.
  • the training vocabulary can include:
  • image categories can include person type and scene type
  • the extracted feature descriptors are clustered by K-Means algorithm to generate a training vocabulary.
  • Building the image library to be retrieved may include:
  • the steps of image retrieval may include:
  • the optional embodiment utilizes the dense distribution of important information on the image and the word bag model for image retrieval, and the extracted features have certain semantics, can describe the content of the image, and highlight the image in the image.
  • the key information reduces some of the secondary information and improves the accuracy of the search.
  • FIG. 5 is a flow chart of a vocabulary training method in accordance with an alternate embodiment of the present disclosure.
  • step 510 a training picture is collected.
  • the image category may include characters, scenes, and the like, and the number of pictures is not too small, and may be greater than a threshold, for example, 200 or 300. The more the number of pictures, the better the processing effect, that is, the larger the threshold, the better the image processing effect.
  • step 520 feature descriptors for all feature points on all pictures are obtained.
  • the step 520 can be implemented by the following steps:
  • the feature points are selected by calculating the positions of all the feature points on the image. For example, some feature points are selected at a small distance from the center of the image, and a small number of feature points are selected at the edge of the image.
  • the manner of calculating the position of all feature points on each image is: specifying an initial feature point having two-dimensional coordinates, according to the coordinates of the previous feature point and the image center coordinate
  • the distance between the steps is calculated, and the position coordinates of the current feature point are calculated according to the step size calculated above, and the position coordinates of the current feature point are recorded.
  • the step size can be determined according to the distance between the coordinates of the previous feature point and the center coordinate of the image, and the smaller the distance, the smaller the step size.
  • the abscissa and the ordinate of the previous feature point are respectively added to the calculated step size to obtain the position of the current feature point.
  • step 530 all the extracted feature descriptors are clustered by the K-Means algorithm to generate a vocabulary.
  • FIG. 6 is a flow chart of a method for establishing an image library according to an alternative embodiment of the present disclosure. As shown in FIG. 6, the method can process each picture in the image library.
  • step 610 the image is pre-processed to normalize the image size.
  • step 620 the position of all feature points on the image is calculated.
  • an initial feature point may be set, the step size is calculated according to the distance of the previous feature point from the center of the image, the position of the current feature point is calculated according to the step size in the above, the position of the current feature point is recorded, and the above manner is cycled until the calculation is performed.
  • the location of all feature points may be set, the step size is calculated according to the distance of the previous feature point from the center of the image, the position of the current feature point is calculated according to the step size in the above, the position of the current feature point is recorded, and the above manner is cycled until the calculation is performed. The location of all feature points.
  • step 630 feature descriptors for all feature points are calculated.
  • step 640 the feature descriptors of the calculated feature points are mapped onto the vocabulary to obtain a histogram.
  • step 650 the sum of the number of occurrences of all words in the histogram is counted.
  • step 660 the number of occurrences of each vocabulary in the histogram is divided by the sum calculated in step 650. Among them, the vocabulary in the vocabulary is generated after the previous "training vocabulary".
  • step 670 the histogram and the associated information of the image calculated in step 660 are stored in a file or database.
  • FIG. 7 is a flow chart of an image retrieval method of an alternative embodiment of the present disclosure.
  • step 710 the image to be retrieved is pre-processed to normalize the image size.
  • step 720 the location of all feature points on the image to be retrieved is calculated.
  • the calculating the position of all the feature points may include setting an initial feature point, calculating the step size according to the distance of the previous feature point from the center of the image, calculating the position of the current feature point according to the step size in the above, and recording the position of the current feature point, Loop through the above process until the position of all feature points is calculated.
  • step 730 feature descriptors for all feature points are calculated.
  • step 740 the feature descriptors of the calculated feature points are mapped onto the vocabulary to obtain a histogram.
  • step 750 the sum of the number of occurrences of all words in the histogram of the image to be retrieved is counted.
  • step 760 the number of occurrences of each vocabulary in the histogram of the image to be retrieved is divided by the sum calculated above.
  • step 770 the similarity of the histogram of the image to be retrieved to each histogram in the image library is calculated.
  • step 780 based on the similarity, and filtering out the picture whose similarity of the histogram of the image to be retrieved is not greater than a threshold.
  • step 790 the results of the search are returned in accordance with the user's request.
  • Embodiments of the present disclosure also provide a non-transitory computer readable storage medium.
  • the foregoing storage medium may be configured to store program code for performing the following steps:
  • Mapping the extracted feature descriptors to a pre-generated vocabulary, and the histories to be retrieved are statistically histograms on the vocabulary
  • the embodiment of the present disclosure further provides a hardware structure diagram of an electronic device.
  • the electronic device includes:
  • At least one processor 80 which is exemplified by a processor 80 in FIG. 8; and a memory 81, may further include a communication interface 82 and a bus 83.
  • the processor 80, the communication interface 82, and the memory 81 can complete communication with each other through the bus 83.
  • Communication interface 82 can be used for information transfer.
  • Processor 30 may invoke logic instructions in memory 31 to perform the methods of the above-described embodiments.
  • logic instructions in the memory 81 described above may be implemented in the form of a software functional unit and sold or used as a stand-alone product, and may be stored in a computer readable storage medium.
  • the memory 81 is used as a computer readable storage medium for storing software programs and computers. Executing a program, such as a program instruction or module corresponding to a method in an embodiment of the present disclosure.
  • the processor 80 executes the functional application and the data processing by executing a software program, an instruction or a module stored in the memory 81, that is, implementing the method in the above method embodiment.
  • the memory 81 may include a storage program area and an storage data area, wherein the storage program area may store an operating system, an application required for at least one function; the storage data area may store data created according to use of the terminal device, and the like. Further, the memory 81 may include a high speed random access memory, and may also include a nonvolatile memory.
  • the above-described modules or steps of the present disclosure may be implemented by a computing device, and the modules or steps of the present disclosure may be centralized on a single computing device or distributed over a network of multiple computing devices. Alternatively, the modules or steps of the present disclosure may be implemented with program code executable by a computing device and stored in a storage device for execution by the computing device.
  • the steps shown or described may be performed in a different order than in the above-described embodiments, or the above modules may be separately fabricated into a plurality of integrated circuit modules, or a plurality of modules or steps in a step or module. It is made into a single integrated circuit module.
  • the image retrieval method and device provided by the present disclosure avoids the phenomenon that the image retrieval method is single and the matching degree is not high in the related art.

Abstract

一种图像的检索方法及装置,其中,该方法包括:提取待检索图像上用于表征图像特征属性的多个特征描述子(110);将提取到的特征描述子映射到预先生成的词汇表上,并统计所述待检索图像在词汇表上的直方图(120);以及计算待检索图像的直方图与预存在图像库中每个图像的直方图的相似度,从图像库中检索出与所述待检索图像的直方图的相似度大于预设阈值的所有图像(130)。

Description

图像的检索方法及装置 技术领域
本公开涉及图像检索领域,例如,涉及一种图像的检索方法及装置。
背景技术
随着网络及照相设备的普及,例如带有照相功能的手机的普及,人们接触到的图像越来越多,因此如何在大量的图像中快速并准确的找到一幅图像的相似图像变得越来越重要。
相关技术中的图像检索技术有基于文本的图像检索技术和基于图像的图像检索技术。其中,基于文本的图像检索技术采用文本检索技术,人工给图像库中的每个图片添加标注,用于描述图像的信息,用户检索时也通过文字来检索一类图片,如百度图片可以支持这种图像检索方法。随着数字图像处理、模式识别以及机器学习等技术的发展,基于图像的图像检索技术应运而生。基于图像的图像检索技术大量应用数字图像处理、模式识别以及机器学习领域的原理和知识,通过特定的算法提取图像的特征,通过提取到的特征计算图像之间的相似度,并依据相似度返回图像库中相似的图片,完成整个图像检索流程。
基于文本的图像检索虽然搜索出来的图片的主题大都是符合用户输入的待检索图片的语义,但基于文本的图像检索方法的图像库中的每张图像都需要人工标注,耗费大量的人力,在互联网图片更新越来越快以及人工成本越来越高的背景下,该方法正变得越来越不实用。相关技术中的基于图像的图像检索系统大都通过计算图像在一个或几个特征上的相似度来检索相近的图片,上述特征包括颜色特征、直方图特征、梯度特征以及几何特征等,但这些特征大都不具有语义性,导致检索出的图片与输入的图片匹配度不高。
针对相关技术中的上述问题,尚未存在有效的解决方案。
发明内容
本公开提供了一种图像的检索方法及装置,避免了相关技术中图像的检索方式单一且匹配度不高的现象。
本公开提供了一种图像的检索方法,包括:提取待检索图像上用于表征图像特征属性的多个特征描述子;将提取到的特征描述子映射到预先生成的词汇表上,并统计在所述词汇表上的直方图;以及计算所述待检索图像的直方图与预存在图像库中每个图像的直方图的相似度,并从所述图像库中检索出与所述待检索图像的直方图的相似度大于预设阈值的所有图像。
可选地,在提取待检索图像上用于表征图像特征属性的多个特征描述子之前,所述方法还包括:采集多种类型的图像,并对所述多种类型的图像进行预处理得到归一化后的多种类型的图像;以及提取归一化后的所述多种类型的图像的特征描述子,并对所述多种类型的图像的特征描述子进行聚类处理生成用于映射多种不同图像特征属性的特征描述子的词汇表。
可选地,在对所述多种类型的图像的特征描述子进行聚类处理生成所述词汇表之后,所述方法还包括:提取归一化后的所述多种类型的图像的特征描述子;将所述多种类型的图像中每个图像上所有的特征描述子映射到词汇表上,并统计出所述每个图像在词汇表上每个词汇的直方图;以及将所述直方图归一化得到所述每张图像在所述词汇表上的归一化后的直方图。
可选地,所述计算所述待检索图像的直方图与预存在图像库中每个图像的直方图的相似度包括:统计所述待检索图像的直方图中每一项代表不同图像特征属性的特征描述子的数量占所述待检索图像的直方图中所有特征描述子总和的比例值;以及根据所述比例值,计算所述待检索图像的直方图与预存在图像库中每个图像的直方图的相似度。
可选地,每一所述直方图中所有不同图像特征属性的特征描述子的比例值的和为1。
本公开的还提供了一种图像的检索装置,包括:特征提取模块,设置为提取待检索图像上用于表征图像特征属性的多个特征描述子;第一映射模块,设置为将提取到的特征描述子映射到预先生成的词汇表上,并统计在所述词汇表上的直方图;以及相似度计算模块,设置为计算所述待检索图像的直方图与预存在图像库中每个图像的直方图的相似度,并从所述图像库中检索出与所述待检索图像的直方图的相似度大于预设阈值的所有图像。
可选地,所述装置还包括:第一处理模块,设置为在提取待检索图像上用于表征图像特征属性的多个特征描述子之前,采集多种类型的图像,并对所述 多种类型的图像进行预处理得到归一化后的所述多种类型的图像;以及第二处理模块,设置为提取归一化后的多种类型的图像的特征描述子,并对所述多种类型的图像的特征描述子进行聚类处理生成用于映射多种不同图像特征属性的特征描述子的词汇表。
可选地,所述装置还包括:提取模块,设置为在对所述多种类型的图像的特征描述子进行聚类处理生成所述词汇表之后,提取归一化后的所述多种类型的图像库中所有图像的特征描述子;第二映射模块,设置为将所述多种类型的图像中的每个图像上所有的特征描述子映射到词汇表上,并统计出所述每个图像在词汇表上每个词汇的直方图;以及归一化模块,设置为将所述直方图归一化得到所述每张图像在所述词汇表上的归一化后的直方图。
可选地,所述相似度计算模块包括:统计单元,设置为统计所述待检索图像的直方图中每一项代表不同图像特征属性的特征描述子的数量占所述待检索图像的直方图中所有特征描述子总和的比例值;以及相似度计算单元,设置为根据所述比例值,计算所述待检索图像的直方图与预存在图像库中每个图像的直方图的相似度。
可选地,每一所述直方图中所有不同图像特征属性的特征描述子的比例值的和为1。
本公开还提供了一种非暂态计算机可读存储介质,存储有计算机可执行指令,所述计算机可执行指令设置为执行上述方法。
本公开还提供了一种电子设备,包括:
至少一个处理器;以及
与所述至少一个处理器通信连接的存储器;其中,
所述存储器存储有可被所述至少一个处理器执行的指令,所述指令被所述至少一个处理器执行,以使所述至少一个处理器执行上述的方法。
本公开中通过提取待检索图像上用于表征图像特征属性的多个特征描述子,将提取到的特征描述子映射到预先生成的词汇表上,并统计所述待检索图像在所述词汇表上的直方图,通过计算待检索图像的直方图与预存在图像库中每个图像的直方图的相似度,并从图像库中检索出与所述待检索图像的直方图的相似度大于预设阈值的所有图像,避免了相关技术中图像的检索方式单一且 匹配度不高的现象。
附图说明
此处所说明的附图用来提供对本公开的理解,构成本申请的一部分,本公开的示意性实施例及实施例的说明用于解释本公开,并不构成对本公开的不当限定。
图1是本公开实施例的图像的检索方法的流程图;
图2是本公开可选实施例的图像的检索装置的结构框图;
图3是本公开可选实施例的图像的检索装置的可选结构框图一;
图4是本公开可选实施例的图像的检索装置的可选结构框图二;
图5是本公开可选实施例的词汇表训练方法的流程图;
图6是本公开可选实施例的图像库建立方法的流程图;
图7是本公开可选实施例的图像检索方法的流程图;以及
图8是本公开实施例提供的电子设备的硬件结构示意图。
具体实施方式
下文中将参考附图并结合实施例来详细说明本公开。在不冲突的情况下,本申请中的实施例及实施例中的特征可以相互组合。
本公开的说明书和权利要求书及上述附图中的术语“第一”、“第二”等是用于区别类似的对象,而不限定特定的顺序或先后次序。
在本实施例中提供了一种图像的检索方法,图1是本公开实施例的图像的检索方法的流程图。
在步骤110中,提取待检索图像上用于表征图像特征属性的多个特征描述子。
在步骤120中,将提取到的特征描述子映射到预先生成的词汇表上,并统计所述待检索图像在所述词汇表上的直方图。
在步骤130中,计算待检索图像的直方图与预存在图像库中每个图像的直方图的相似度,并从图像库中检索出与所述待检索图像的直方图的相似度大于 预设阈值的所有图像。
通过本实施例的上述步骤110至步骤130可知,获取待检索图像上用于表征图像特征属性的多个特征描述子,将提取到的特征描述子映射到预先生成的词汇表上,并统计所述待检索图像在所述词汇表上的直方图,通过计算待检索图像的直方图与预存在图像库中每个图像的直方图的相似度,并从图像库中检索出与所述待检索图像的直方图的相似度大于预设阈值的所有图像,即,通过将图像的特征描述子映射到词汇表中并通过直方图来表示不同类型的图像特征属性,通过比较待检索图像的直方图与预存的直方图,将与待检索图像的直方图的相似度高于预设阈值的图像检索出来,避免了相关技术中图像的检索方式单一且匹配度不高的现象。
在本实施例的可选实施方式中,在步骤110中的获取待检索图像上用于表征图像特征属性的特征描述子之前,本实施例的方法还可以包括:
采集多种类型的图像,并对多种类型的图像进行预处理得到归一化后的多种类型的图像;以及
提取归一化后的多种类型的图像的特征描述子,并对多种类型的图像的特征描述子进行聚类处理生成用于映射多种不同图像特征属性的特征描述子的词汇表。
例如,在以下场景中:搜集训练图像,该图像可以包括人物和景物两种类型的图像;对搜集到的图片进行预处理,使图像尺寸大小归一化,利用重要信息的稠密分布的特点提取每个训练图像上对应的特征点的特征描述子,对提取到的特征描述子进行聚类处理后生成用于表征图像特征属性的词汇表。特征点的特征描述子的重要信息可以表征该特征描述子的重要程度,当特征点的位置到图像中心的距离越小时,特征点的特征描述子越重要。
此外,在本实施例的一个可选实施方式中,在对所述多种类型的图像的特征描述子进行聚类处理生成所述词汇表之后,本实施例的方法还可以包括:
提取归一化后的所述多种类型的图像的特征描述子;
将所述多种类型的图像中的每个图像上所有的特征描述子映射到词汇表上,并统计出该每个图像在词汇表上每个词汇的直方图;以及
将所述直方图归一化得到每张图像在所述词汇表上的归一化后的直方图。
例如,对图像库中的图片进行预处理,使图像尺寸大小归一化;利用重要信息稠密分布的特点提取每个图像上对应点的特征描述子;将提取到的特征描述子映射到预先生成的词汇表上,并统计所述待检索图像在所述词汇表上的直方图;统计直方图中所有特征描述子出现次数的总和,将直方图中的每一项代表不同图像特征属性的特征描述子的出现次数都除以该总和,得到比例值,确保处理后的直方图每项代表不同特征属性的特征描述子的比例值之和为1,将处理后的直方图数据保存到文件或数据库中。
对于本实施例中的步骤130,计算待检索图像的直方图与预存在图像库中所有图像的直方图的相似度包括:
统计待检索图像的直方图中每一项代表不同图像特征属性的特征描述子的数量占所述待检索图像的直方图中所有特征描述子总和的比例值;
根据所述比例值,计算待检索图像的直方图与预存在图像库中每个图像的直方图的相似度。
其中,本实施例中的每一直方图中所有不同图像特征属性的特征描述子的比例值的和为1。
通过以上的实施方式的描述,本领域的技术人员可以清楚地了解到上述实施例的方法可借助软件加硬件平台的方式来实现,当然也可以通过硬件实现。本公开的技术方案本质上可以以软件产品的形式体现出来,该计算机软件产品存储在一个存储介质(如只读存储器(Read-only Memory,ROM)、随机存储存储器(Random-Access Memory,RAM)、磁碟、光盘)中,包括一个或多个指令用以使得一台终端设备(可以是手机,计算机,服务器,或者网络设备等)执行本公开实施例的方法。
在本实施例中还提供了一种图像的检索装置,该装置可实现上述实施例及可选实施方式。如以下所使用的,术语“模块”可以实现预定功能的软件和硬件中至少一种的组合。尽管以下实施例所描述的装置可以用软件来实现,但是也可以用硬件,或者软件和硬件的组合的方式实现。
图2是本公开可选实施例的图像的检索装置的结构框图,如图2所示,该装置包括:特征提取模块22、第一映射模块24和相似度计算模块26。
特征提取模块22设置为提取待检索图像上用于表征图像特征属性的多个特 征描述子。第一映射模块24,与特征提取模块22耦合连接,设置为将提取到的特征描述子映射到预先生成的词汇表上,并统计所述待检索图像在所述词汇表上的直方图。相似度计算模块26,与第一映射模块24耦合连接,设置为计算待检索图像的直方图与预存在图像库中每个图像的直方图的相似度,并从图像库中检索出与所述待检索图像的直方图的相似度大于预设阈值的所有图像。
图3是本公开可选实施例的图像的检索装置的可选结构框图一,如图3所示,该装置还可以包括:第一处理模块32和第二处理模块34。第一处理模块32设置为在提取待检索图像上用于表征图像特征属性的多个特征描述子之前,采集多种类型的图像,并对多种类型的图像进行预处理得到归一化后的多种类型的图像。第二处理模块34,与第一处理模块32耦合连接,设置为提取归一化后的多种类型的图像的特征描述子,并对多种类型的图像的特征描述子进行聚类处理生成用于映射多种不同图像特征属性的特征描述子的词汇表。
图4是本公开可选实施例的图像的检索装置的可选结构框图二,如图4所示,该装置还可以包括:提取模块42、第二映射模块44和归一化模块46。提取模块42设置为在对所述多种类型的图像的特征描述子进行聚类处理生成所述词汇表之后,提取归一化后的所述多种类型的图像的特征描述子。第二映射模块44,与提取模块42耦合连接,设置为将所述多种类型的图像中的每个图像上所有的特征描述子映射到词汇表上,并统计出每个图像在词汇表上每个词汇的直方图。归一化模块46,与第二映射模块44耦合连接,设置为将所述直方图归一化得到每张图像在所述词汇表上的归一化后的直方图。
可选地,本实施例图2中的相似度计算模块26还可以包括:统计单元和相似度计算单元。统计单元设置为统计待检索图像的直方图中每一项代表不同图像特征属性的特征描述子的数量占所述待检索图像的直方图中所有特征描述子总和的比例值。相似度计算单元设置为根据所述比例值,计算待检索图像的直方图与预存在图像库中每个图像的直方图的相似度。
可选地,在本实施例中每一直方图中所有不同图像特征属性的特征描述子的比例值的和为1。
上述模块是可以通过软件或硬件来实现的,当采用硬件来实现时,上述模块均位于同一处理器中,或者,上述模块分别位于多个处理器中。
本公开提出了一种基于重要信息稠密分布特点的词袋模型图像检索方法, 本可选实施例通过提取图像的稠密特征,通过聚类处理生成一定量的具有代表性的特征描述子,组成词汇表。在检索阶段,通过衡量待检索图片与图像库中的每个图片在词汇表上的直方图的相似度来完成图像检索。在本可选实施例中涉及到的词袋模型可以提取一定的语义信息,利用图像上重要信息的稠密分布特点可以提取到更多的重要信息,忽略掉一定的次要信息。
其中,所述稠密特征可以是指利用重要信息在图像上的稠密分布特点(即图像中心附近的重要信息较多,可以在图像中心附近多选择一些特征点)选取特征点,并提取选取的特征点的特征描述子。通过上述方法提取到的所有特征点及特征描述子叫做图像的稠密特征。
本可选实施例的检索方法可以包括训练词汇表、构建待检索图像库以及图像检索。
训练词汇表可以包括:
搜集训练图片,图像类别可以包括人物类型和景物类型等;
对搜集到的图片进行预处理,使图像尺寸大小归一化;
利用重要信息稠密分布的特点提取每个训练图像上对应点的特征描述子;以及
对提取到的特征描述子采用K-平均(K-Means)算法进行聚类,生成训练词汇表。
构建待检索图像库可以包括:
对图像库中的图片进行预处理,使图像尺寸大小归一化;
利用重要信息的稠密分布特点提取每个图像上对应点的特征描述子;
将提取到的特征映射到预先生成的词汇表上,并统计每个图像在所述词汇表上的直方图;
统计直方图中所有特征描述子出现次数的总和,将直方图中的每一项代表不同图像特征属性的特征描述子的数量都除以该总和,确保处理后的直方图每项代表不同图像特征属性的特征描述子的比例值之和为1;以及
将处理后的直方图数据保存到文件或数据库中。
图像检索的步骤可以包括:
对待检索图片进行预处理,使图像尺寸大小归一化;
利用重要信息稠密分布的特点提取待检索图像上对应点的特征描述子;
将提取到的特征映射到预先生成的词汇表上,并统计直方图;
统计直方图中项出现次数的总和,将直方图中的每一项词汇出现次数的都除以该总和,确保处理后的直方图每项属性的特征描述子的比例值之和为1;
计算待检索图像的直方图与图像库中每张图像预先计算好的直方图之间的相似度;
对相似度进行排序,并获取相似度高于一阈值的所有图像的信息;以及
按用户要求,返回最相似的k张图片。
由此可知,本可选实施例利用图像上重要信息稠密分布的特点以及词袋模型进行图像检索,所提取到的特征既具有一定的语义性,可以描述图像的内容,又突出了图像中的重点信息,减少了一部分次要信息,提高了检索的准确度。
图5是根据本公开可选实施例的词汇表训练方法的流程图。
在步骤510中,搜集训练图片。
其中,图像类别可以包括人物和景物等,图片数目不太少,可以大于一阈值,例如是200或者300。图片数目越多,处理效果越好,即阈值越大,图像处理效果一般越好。
在步骤520中,得到所有图片上所有特征点的特征描述子。
其中,该步骤520可以通过如下步骤来实现:
对图片进行预处理,使图像尺寸大小归一化;
计算图像上所有特征点的位置;以及
计算所有特征点的特征描述子。
上述步骤中,通过计算图像上所有特征点的位置来选取特征点,比如,在与图像中心距离较小位置多选取一些特征点,在图像的边缘地带选取少量的特征点。
其中,该计算每张图像上所有特征点的位置的可以方式为:指定一个初始特征点,该初始特征点具有二维坐标,根据前一特征点的坐标与图像中心坐标 之间的距离计算步长,依据上述计算得到的步长计算当前特征点的位置坐标,记录当前特征点的位置坐标。步长可以根据前一特征点的坐标与图像中心坐标之间的距离确定,距离越小步长越小。前一个特征点的横坐标和纵坐标分别与计算出的步长相加,得到当前特征点的位置。
在步骤530中,对提取到的所有特征描述子通过K-Means算法进行聚类,生成词汇表。
图6是是本公开可选实施例的图像库建立方法的流程图,如图6所示,该方法可以处理图像库中的每张图片。
在步骤610中,对图像进行预处理,使图像尺寸大小归一化。
在步骤620中,计算图像上所有特征点的位置。
其中,可以设置一个初始特征点,根据前一特征点距图像中心的距离计算步长,依据上述中的步长计算当前特征点的位置,记录当前特征点的位置,循环上述的方式直到计算出所有特征点的位置。
在步骤630中,计算所有特征点的特征描述子。
在步骤640中,将计算的特征点的特征描述子映射到词汇表上,得到直方图。
在步骤650中,统计直方图中所有词汇出现的次数的总和。
在步骤660中,将直方图中每个词汇出现的次数除以步骤650中计算出的总和。其中,词汇表中的词汇是前面“训练词汇表”后生成的。
在步骤670中,将步骤660中计算的直方图及图像的相关信息存文件或数据库。
图7是本公开可选实施例的图像检索方法的流程图。
在步骤710中,对待检索图像进行预处理,使图像尺寸大小归一化。
在步骤720中,计算待检索图像上所有特征点的位置。
其中,计算所有特征点的位置可以包括设置一个初始特征点,根据前一特征点距图像中心的距离计算步长,依据上述中的步长计算当前特征点的位置,记录当前特征点的位置,循环上述过程直到计算出所有特征点的位置。
在步骤730中,计算所有特征点的特征描述子。
在步骤740中,将计算的特征点的特征描述子映射到词汇表上,得到直方图。
在步骤750中,统计待检索图像的直方图中所有词汇出现的次数的总和。
在步骤760中,将待检索图像的直方图中每个词汇出现的次数除以上述计算出的总和。
在步骤770中,计算待检索图像的直方图与图像库中的每个直方图的相似度。
在步骤780中,依据相似度,并用将于待检索图像的直方图的相似度不大于一个阈值的图片过滤掉。
在步骤790中,依据用户的要求,返回检索的结果。
本公开的实施例还提供了一种非暂态计算机可读存储介质。可选地,在本实施例中,上述存储介质可以被设置为存储用于执行以下步骤的程序代码:
提取待检索图像上用于表征图像特征属性的多个特征描述子;
将提取到的特征描述子映射到预先生成的词汇表上,并所述待检索图像统计在词汇表上的直方图;以及
计算待检索图像的直方图与预存在图像库中每个图像的直方图的相似度,并从图像库中检索出与所述待检索图像的直方图的相似度大于预设阈值的所有图像。
本公开实施例还提供了一种电子设备的硬件结构示意图。参见图8,该电子设备包括:
至少一个处理器(processor)80,图8中以一个处理器80为例;和存储器(memory)81,还可以包括通信接口(Communications Interface)82和总线83。其中,处理器80、通信接口82、存储器81可以通过总线83完成相互间的通信。通信接口82可以用于信息传输。处理器30可以调用存储器31中的逻辑指令,以执行上述实施例的方法。
此外,上述的存储器81中的逻辑指令可以通过软件功能单元的形式实现并作为独立的产品销售或使用时,可以存储在一个计算机可读取存储介质中。
存储器81作为一种计算机可读存储介质,可用于存储软件程序、计算机可 执行程序,如本公开实施例中的方法对应的程序指令或模块。处理器80通过运行存储在存储器81中的软件程序、指令或模块,从而执行功能应用以及数据处理,即实现上述方法实施例中的方法。
存储器81可包括存储程序区和存储数据区,其中,存储程序区可存储操作系统、至少一个功能所需的应用程序;存储数据区可存储根据终端设备的使用所创建的数据等。此外,存储器81可以包括高速随机存取存储器,还可以包括非易失性存储器。上述的本公开的模块或步骤可以用计算装置来实现,本公开的模块或步骤可以集中在单个的计算装置上,或者分布在多个计算装置所组成的网络上。可选地,本公开的模块或步骤可以用计算装置可执行的程序代码来实现,并存储在存储装置中由计算装置来执行。在一些情况下,可以以不同于此上述实施例中的顺序执行所示出或描述的步骤,或者将上述模块分别制作成多个集成电路模块,或者将步骤或模块中的多个模块或步骤制作成单个集成电路模块来实现。
工业实用性
本公开提供的图像的检索方法及装置,避免了相关技术中图像的检索方式单一且匹配度不高的现象。

Claims (11)

  1. 一种图像的检索方法,包括:
    提取待检索图像上用于表征图像特征属性的多个特征描述子;
    将提取到的特征描述子映射到预先生成的词汇表上,并统计所述待检索图像在所述词汇表上的直方图;以及
    计算所述待检索图像的直方图与预存在图像库中每个图像的直方图的相似度,并从所述图像库中检索出与所述待检索图像的直方图的相似度大于预设阈值的所有图像。
  2. 根据权利要求1所述的方法,在提取待检索图像上用于表征图像特征属性的多个特征描述子之前,所述方法还包括:
    采集多种类型的图像,并对所述多种类型的图像进行预处理得到归一化后的多种类型的图像;以及
    提取归一化后的多种类型的图像的特征描述子,并对所述多种类型的图像的特征描述子进行聚类处理生成用于映射多种不同图像特征属性的特征描述子的词汇表。
  3. 根据权利要求2所述的方法,在对所述多种类型的图像的特征描述子进行聚类处理生成所述词汇表之后,所述方法还包括:
    提取归一化后的所述多种类型的图像的特征描述子;
    将所述多种类型的图像中的每个图像上所有的特征描述子映射到词汇表上,并统计出所述每个图像在词汇表上每个词汇的直方图;以及
    将所述直方图归一化得到所述每张图像在所述词汇表上的归一化后的直方图。
  4. 根据权利要求3所述的方法,其中,所述计算所述待检索图像的直方图与预存在图像库中每个图像的直方图的相似度包括:
    统计所述待检索图像的直方图中每一项代表不同图像特征属性的特征描述子的数量占所述待检索图像的直方图中所有特征描述子总和的比例值;以及
    根据所述比例值,计算所述待检索图像的直方图与预存在图像库中每个图像的直方图的相似度。
  5. 根据权利要求4所述的方法,其中,每一所述直方图中所有不同图像特征属性的特征描述子的比例值的和为1。
  6. 一种图像的检索装置,包括:
    特征提取模块,设置为提取待检索图像上用于表征图像特征属性的多个特征描述子;
    第一映射模块,设置为将提取到的特征描述子映射到预先生成的词汇表上,并统计在所述词汇表上的直方图;以及
    相似度计算模块,设置为计算所述待检索图像的直方图与预存在图像库中每个图像的直方图的相似度,并从所述图像库中检索出与所述待检索图像的直方图的相似度大于预设阈值的所有图像。
  7. 根据权利要求6所述的装置,还包括:
    第一处理模块,设置为在提取待检索图像上用于表征图像特征属性的多个特征描述子之前,采集多种类型的图像,并对所述多种类型的图像进行预处理得到归一化后的多种类型的图像;以及
    第二处理模块,设置为提取归一化后的多种类型的图像的特征描述子,并对所述多种类型的图像的特征描述子进行聚类处理生成用于映射多种不同图像特征属性的特征描述子的词汇表。
  8. 根据权利要求7所述的装置,还包括:
    提取模块,设置为在对所述多种类型的图像的特征描述子进行聚类处理生 成所述词汇表之后,提取归一化后的所述多种类型的图像的特征描述子;
    第二映射模块,设置为将所述多种类型的图像中的每个图像上所有的特征描述子映射到词汇表上,并统计出所述每个图像在词汇表上每个词汇的直方图;以及
    归一化模块,设置为将所述直方图归一化得到所述每张图像在所述词汇表上的归一化后的直方图。
  9. 根据权利要求8所述的装置,其中,所述相似度计算模块包括:
    统计单元,设置为统计所述待检索图像的直方图中每一项代表不同图像特征属性的特征描述子的数量占所述待检索图像的直方图中所有特征描述子总和的比例值;以及
    相似度计算单元,设置为根据所述比例值,计算所述待检索图像的直方图与预存在图像库中每个图像的直方图的相似度。
  10. 根据权利要求9所述的装置,其中,每一所述直方图中所有不同图像特征属性的特征描述子的比例值的和为1。
  11. 一种非暂态计算机可读存储介质,存储有计算机可执行指令,所述计算机可执行指令设置为执行权利要求1-5中任一项的方法。
PCT/CN2017/074356 2016-02-22 2017-02-22 图像的检索方法及装置 WO2017143979A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201610097309.6A CN107103002A (zh) 2016-02-22 2016-02-22 图像的检索方法及装置
CN201610097309.6 2016-02-22

Publications (1)

Publication Number Publication Date
WO2017143979A1 true WO2017143979A1 (zh) 2017-08-31

Family

ID=59658575

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2017/074356 WO2017143979A1 (zh) 2016-02-22 2017-02-22 图像的检索方法及装置

Country Status (2)

Country Link
CN (1) CN107103002A (zh)
WO (1) WO2017143979A1 (zh)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108764258B (zh) * 2018-05-24 2022-03-04 西安电子科技大学 一种用于群体图像插入的最优图像集选取方法
CN110059212A (zh) * 2019-03-16 2019-07-26 平安科技(深圳)有限公司 图像检索方法、装置、设备及计算机可读存储介质

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101136015A (zh) * 2006-09-01 2008-03-05 北大方正集团有限公司 一种计算图像之间相似度的方法
CN102930302A (zh) * 2012-10-18 2013-02-13 山东大学 基于在线序贯极限学习机的递增式人体行为识别方法
CN103064985A (zh) * 2013-01-28 2013-04-24 北京理工大学 基于先验知识的图像检索方法
CN103278170A (zh) * 2013-05-16 2013-09-04 东南大学 基于显著场景点检测的移动机器人级联地图创建方法
CN105069809A (zh) * 2015-08-31 2015-11-18 中国科学院自动化研究所 一种基于平面混合标识物的相机定位方法及系统

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8345102B2 (en) * 2009-01-13 2013-01-01 Futurewei Technologies, Inc. Image processing system and method for object tracking
CN103324650A (zh) * 2012-10-23 2013-09-25 深圳市宜搜科技发展有限公司 一种图像检索方法及系统
CN103593464B (zh) * 2013-11-25 2017-02-15 华中科技大学 基于视觉特征的视频指纹检测及视频序列匹配方法及系统
CN105183746B (zh) * 2015-07-08 2018-04-17 西安交通大学 从多相关图片中挖掘显著特征实现图像检索的方法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101136015A (zh) * 2006-09-01 2008-03-05 北大方正集团有限公司 一种计算图像之间相似度的方法
CN102930302A (zh) * 2012-10-18 2013-02-13 山东大学 基于在线序贯极限学习机的递增式人体行为识别方法
CN103064985A (zh) * 2013-01-28 2013-04-24 北京理工大学 基于先验知识的图像检索方法
CN103278170A (zh) * 2013-05-16 2013-09-04 东南大学 基于显著场景点检测的移动机器人级联地图创建方法
CN105069809A (zh) * 2015-08-31 2015-11-18 中国科学院自动化研究所 一种基于平面混合标识物的相机定位方法及系统

Also Published As

Publication number Publication date
CN107103002A (zh) 2017-08-29

Similar Documents

Publication Publication Date Title
US10599709B2 (en) Object recognition device, object recognition method, and program for recognizing an object in an image based on tag information
WO2019218514A1 (zh) 网页目标信息的提取方法、装置及存储介质
EP3028184B1 (en) Method and system for searching images
CN107463605B (zh) 低质新闻资源的识别方法及装置、计算机设备及可读介质
US11727053B2 (en) Entity recognition from an image
US8577882B2 (en) Method and system for searching multilingual documents
WO2019041521A1 (zh) 用户关键词提取装置、方法及计算机可读存储介质
US20100250539A1 (en) Shape based picture search
CN111062871A (zh) 一种图像处理方法、装置、计算机设备及可读存储介质
CN109582813B (zh) 一种文物展品的检索方法、装置、设备和存储介质
WO2021237570A1 (zh) 影像审核方法及装置、设备、存储介质
US20160253577A1 (en) Image Clustering Method, Image Clustering System, And Image Clustering Server
WO2020134684A1 (zh) 信息检索方法、装置、设备和介质
WO2016101766A1 (zh) 相似人脸图片获取和人脸图片信息获取方法和装置
CN107291825A (zh) 一种视频中同款商品的检索方法和系统
CN113434716B (zh) 一种跨模态信息检索方法和装置
CN106844381B (zh) 图像处理装置及方法
WO2017143979A1 (zh) 图像的检索方法及装置
CN113221918B (zh) 目标检测方法、目标检测模型的训练方法及装置
JP6017277B2 (ja) 特徴ベクトルの集合で表されるコンテンツ間の類似度を算出するプログラム、装置及び方法
CN110209895B (zh) 向量检索方法、装置和设备
CN102760127A (zh) 基于扩展文本信息来确定资源类型的方法、装置及设备
WO2020083344A1 (zh) 关联确定方法和装置、推荐方法和装置、电子设备、计算机可读存储介质
WO2023035535A1 (zh) 语义分割网络的训练方法、装置、设备及存储介质
CN114860991A (zh) 一种短视频去重方法及计算机可读存储介质

Legal Events

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

Ref country code: DE

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

Ref document number: 17755808

Country of ref document: EP

Kind code of ref document: A1

122 Ep: pct application non-entry in european phase

Ref document number: 17755808

Country of ref document: EP

Kind code of ref document: A1