WO2015024383A1 - Similarity acquisition method for colour distribution and texture distribution image retrieval - Google Patents

Similarity acquisition method for colour distribution and texture distribution image retrieval Download PDF

Info

Publication number
WO2015024383A1
WO2015024383A1 PCT/CN2014/073995 CN2014073995W WO2015024383A1 WO 2015024383 A1 WO2015024383 A1 WO 2015024383A1 CN 2014073995 W CN2014073995 W CN 2014073995W WO 2015024383 A1 WO2015024383 A1 WO 2015024383A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
distribution
template
pixels
feature
Prior art date
Application number
PCT/CN2014/073995
Other languages
French (fr)
Chinese (zh)
Inventor
徐滢
Original Assignee
成都品果科技有限公司
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 成都品果科技有限公司 filed Critical 成都品果科技有限公司
Publication of WO2015024383A1 publication Critical patent/WO2015024383A1/en

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/40—Analysis of texture
    • 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
    • G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
    • G06V10/75—Organisation of the matching processes, e.g. simultaneous or sequential comparisons of image or video features; Coarse-fine approaches, e.g. multi-scale approaches; using context analysis; Selection of dictionaries
    • G06V10/758—Involving statistics of pixels or of feature values, e.g. histogram matching
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10024—Color image
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
    • G06V10/467—Encoded features or binary features, e.g. local binary patterns [LBP]

Definitions

  • the present invention relates to image retrieval techniques, and more particularly to a similarity acquisition method for color distribution and texture distribution image retrieval. Background technique
  • the technical problem to be solved by the present invention is: To solve the above problems, a similarity acquisition method suitable for cloud storage color distribution and texture distribution image retrieval is provided.
  • Step 1 extracting the color distribution feature and the texture distribution feature of the input image
  • Step 2 respectively calculating the similarity between the color distribution feature of the input image and the color distribution feature of each image in the database, and obtaining an input image and a database
  • Step 201 Convert an image to a color space to obtain an image I;
  • Step 202 Mapping H, S, and components of each pixel of the image into color feature values G:
  • Step 203 Statistic value distribution of each pixel in the statistical image: traverse the color feature value of each pixel, and the statistics fall into each color
  • the number of pixels in the distribution histogram interval is divided by the number of pixels in each color distribution histogram interval divided by the total number of image pixels to obtain a normalized color distribution feature ⁇ t(x), where X represents the color distribution histogram Diagram interval.
  • the method for acquiring the color distribution feature further includes:
  • the method for obtaining the texture distribution feature includes: Step 301: Convert an image into a grayscale image to obtain an image L;
  • Step 302 traversing the image to obtain an LHP feature of each template by using a template with a size of 3 pixels and 3 pixels, wherein the method for obtaining the template L B p feature includes:
  • the gray value of the nine pixels in the template is P' ( Q ⁇ z ⁇ 8 ), where the gray value of the pixel in the middle of the template is recorded as; the gray value of the other pixels in the template is subtracted.
  • Step 303 Obtaining a rotation-invariant feature of each template; wherein the method of obtaining the rotation-invariant LBP feature of the template includes:
  • Step 304 Counting the distribution of each of the rotation-invariant features in each template: traversing each rotation-invariant feature value of each template, and counting the number of pixels falling into each texture distribution histogram interval, and then falling into The number of pixels in each texture distribution histogram interval is divided by the total number of image pixels, and the normalized texture distribution feature hist(y) is obtained, where _y represents the texture distribution histogram interval.
  • the method for calculating the color distribution feature similarity Sa in the step 2 includes: Step 401: Using the formula
  • the method for calculating the texture distribution feature similarity Sb in the step 2 includes:
  • ⁇ is the texture distribution feature of the first image and the texture distribution feature of the second image.
  • Wa Wb.
  • the present invention also protects a similarity acquisition method for a texture distribution image retrieval method, including:
  • Step 1 extracting a texture distribution feature of the input image
  • Step 2 Calculate the similarity between the texture distribution feature of the input image and the texture distribution feature of each image in the database, and obtain the texture distribution feature similarity Sb between the input image and each image in the database (i) ), 1 takes 0, 1, 2... total number of database images -1; the method for obtaining the texture distribution features includes:
  • Step 301 Converting an image into a grayscale image to obtain an image L;
  • Step 302 traversing the image to obtain an LHP feature of each template by using a template with a size of 3 pixels and 3 pixels, wherein the method for obtaining the template L B p feature includes: recording a gray value of 9 pixels in the template P' ( Q ⁇ z ⁇ 8 ), where the gray value of the pixel in the middle of the template is recorded as; the gray value of the other pixels in the template is subtracted.
  • the method for obtaining the template L B p feature includes: recording a gray value of 9 pixels in the template P' ( Q ⁇ z ⁇ 8 ), where the gray value of the pixel in the middle of the template is recorded as; the gray value of the other pixels in the template is subtracted.
  • Step 303 Obtaining a rotation-invariant feature of each template; wherein the method of obtaining the rotation-invariant LBP feature of the template includes:
  • LBP" (i) m [ n (ROR(LBP(il q)) , where 1 ⁇ 8, ROR represents the shift operation, and g 0 represents the number of shift bits;
  • Step 304 Counting the distribution of each of the rotation-invariant features in each template: traversing each rotation-invariant feature value of each template, and counting the number of pixels falling into each texture distribution histogram interval, and then falling into The number of pixels in each texture distribution histogram interval is divided by the total number of image pixels, and the normalized texture distribution feature hist(y) is obtained, where _y represents the texture distribution histogram interval.
  • the present invention also protects the method of acquiring texture distribution features in the above.
  • the above technical solution is adopted, and the beneficial effects of the present invention are as follows:
  • the image similarity obtaining method according to the present invention does not need to make any assumptions on the image, and does not require a large number of labeled sample training models, and is easy to implement.
  • FIG. 1 is a flow chart of color distribution feature extraction in the present invention.
  • the present invention provides a similarity acquisition method for color distribution and texture distribution image retrieval, and the specific steps include:
  • Step 1 extracting the color distribution feature and the texture distribution feature of the input image
  • Step 2 respectively calculating the similarity between the color distribution feature of the input image and the color distribution feature of each image in the database, and obtaining an input image and a database
  • a method for acquiring a color distribution feature includes: Step 201: Convert an image into a color space to obtain an image I; generally, the image is an RGB color space, and the image of the RGB color space is obtained.
  • the conversion to the color space is a technique well known in the art, and the specific process thereof will not be described herein.
  • Step 203 Statistic value distribution of each pixel in the statistical image: traverse the color feature values of each pixel, and count the number of pixels falling into the histogram interval of each color distribution, The number of pixels falling into the histogram interval of each color distribution is divided by the total number of image pixels, respectively, to obtain a normalized color distribution feature ⁇ t(x), where X represents a color distribution histogram interval.
  • the method for acquiring the color distribution feature further includes:
  • the image is divided into N blocks, for example, N is equal to 36; the pixels contained in the image boundary block are counted only once, and the remaining pixels are counted twice.
  • N is equal to 36; the pixels contained in the image boundary block are counted only once, and the remaining pixels are counted twice.
  • collecting the feature value distribution of each pixel in the image traversing the feature value of each pixel, and counting the number of pixels falling into the histogram interval of each color distribution; The pixel points in the image boundary block are counted twice, gp.
  • the number of pixels falling into the interval is increased by 2;
  • the number of pixels falling into the interval is increased by 1; finally, the pixel falling into the histogram interval of each color distribution is added.
  • the number is divided by the total number of image pixels to obtain a normalized color distribution characteristic tet(x), where X represents a color histogram.
  • the method for obtaining the texture distribution feature includes:
  • Step 301 Converting an image into a grayscale image to obtain an image L;
  • ⁇ 0.299 ⁇ + 0.587 *(? + 0.114 R represents the red component of the pixel, G represents the green component of the pixel, and ⁇ represents the blue component of the pixel.
  • 0.299, 0.587, 0.114 are coefficients, although this coefficient is not unique and cannot be construed as limiting the invention.
  • Step 302 traversing the image to obtain a ⁇ feature (ie, a texture feature) of each template by using a template with a size of 3 pixels and 3 pixels, wherein the method for obtaining the template feature includes: recording 9 pixels in the template The gray value is P' ( Q ⁇ z ⁇ 8 ), where the image in the middle of the template The prime gray value is recorded as; the gray value of other pixels in the template is subtracted. get:
  • step 303 obtaining the rotation-invariant LBP" feature of each template;
  • the method of rotating the invariant LBP "features includes: For each H of the template (0 is shifted by operation, respectively, 8 binary data can be obtained, and the smallest one is taken as the rotation-invariant LBP ".
  • LBP" (i) m [ n (ROR(LBP(il q)) , where 1 ⁇ 8, ROR represents the shift operation, and g 0 represents the number of shift bits;
  • Step 304 Counting the distribution of each of the rotation-invariant features in each template: traversing each rotation-invariant feature value of each template, and counting the number of pixels falling into each texture distribution histogram interval, and then falling into The number of pixels in each texture distribution histogram interval is divided by the total number of image pixels, and the normalized texture distribution feature hist(y) is obtained, where _y represents the texture distribution histogram interval.
  • the definition of the texture distribution histogram here is similar to the definition of the aforementioned color distribution histogram.
  • the texture distribution histogram divides the texture features of the entire image into several intervals, and then describes the proportion of different textures in the entire image by the case where each pixel is distributed in each interval.
  • Step 401 Using the formula
  • a specific implementation manner of calculating the texture distribution feature similarity Sb of the two images includes:
  • J represents the texture distribution histogram interval.
  • Step 2 respectively calculating the similarity between the texture distribution feature of the input image and the texture distribution feature of each image in the database, and obtaining a texture distribution feature similarity Sb between the input image and each image in the database (i) ), 1 takes 0, 1, 2. 2. The total number of database images -1. Similarly, when the texture distribution feature similarity between the input image and each image of the database is obtained, each similarity is sorted. The greater the similarity, the more similar the two images are. We can set a threshold according to experience. , will output the image in all databases larger than the combined similarity greater than the threshold as the search result.
  • the invention is not limited to the specific embodiments described above. The invention extends to any novel features or any new combinations disclosed in this specification, as well as a novel or a novel combination of any new method or process disclosed.

Landscapes

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

Abstract

Disclosed is a similarity acquisition method for colour distribution and texture distribution image retrieval, which relates to the technology of image retrieval. The key technical points of the present invention comprise: extracting a colour distribution feature and a texture distribution feature of an input image; respectively calculating the similarity between the colour distribution feature of the input image and a colour distribution feature of each image in a database, so as to obtain the colour distribution feature similarity Sa (i) between the input image and each image in the database; respectively calculating the similarity between the texture distribution feature of the input image and a texture distribution feature of each image in the database, so as to obtain the texture distribution feature similarity Sb (i) between the input image and each image in the database; and using the formula S (i) = Wa * Sa (i) + Wb * Sb (i) to calculate the combined similarity S (i) between the input image and each image in the database.

Description

说 明 书 用于颜色分布和纹理分布图像检索的相似度获取方法 技术领域  Description Method for obtaining similarity for color distribution and texture distribution image retrieval
本发明涉及图像检索技术，尤其是一种用于颜色分布和纹理分布 图像检索的相似度获取方法。 背景技术  The present invention relates to image retrieval techniques, and more particularly to a similarity acquisition method for color distribution and texture distribution image retrieval. Background technique
近年来， 随着移动互联网的迅速发展， 拍照应用获得了很大的发 展空间， 照片的获取与存储变得十分容易。 随着照片数据***式的增 长， 用户迫切需要对照片的检索和整理的自动化技术。现有的图像检 索技术都要依赖数据库中已存图像的训练样本求得相似度。而目前云 存储的照片基本上都是来自各种用户拍摄的各种场景的照片，并没有 可获取的显示标注了的训练样本。因而现有的图像检索技术不便直接 应用到云存储图像的检索中。 发明内容  In recent years, with the rapid development of the mobile Internet, the photo application has gained a lot of development space, and the acquisition and storage of photos has become very easy. With the explosive growth of photo data, users are in urgent need of automated techniques for the retrieval and organization of photos. Existing image retrieval techniques rely on training samples of stored images in the database to obtain similarities. At present, the photos stored in the cloud are basically photos from various scenes taken by various users, and there is no training sample that can be obtained by displaying the display. Therefore, the existing image retrieval technology is inconvenient to directly apply to the retrieval of cloud storage images. Summary of the invention
本发明所要解决的技术问题是： 针对上述存在的问题， 提供一种 适用于云存储颜色分布和纹理分布图像检索的相似度获取方法。  The technical problem to be solved by the present invention is: To solve the above problems, a similarity acquisition method suitable for cloud storage color distribution and texture distribution image retrieval is provided.
本发明提供的用于颜色分布和纹理分布图像检索的相似度获取方 法， 其特征在于， 包括：  The method for acquiring similarity for color distribution and texture distribution image retrieval provided by the present invention is characterized in that it comprises:
歩骤 1 : 提取输入图像的颜色分布特征及纹理分布特征； 歩骤 2: 分别计算所述输入图像的颜色分布特征与数据库中每一 幅图像的颜色分布特征的相似度，得到输入图像与数据库中每一幅图 像之间的颜色分布特征相似度 Sa (0, i取 0、 1、 2...数据库图像总 数 -1 ;  Step 1: extracting the color distribution feature and the texture distribution feature of the input image; Step 2: respectively calculating the similarity between the color distribution feature of the input image and the color distribution feature of each image in the database, and obtaining an input image and a database The color distribution characteristic similarity Sa between each image in each case (0, i takes 0, 1, 2... total number of database images -1;
分别计算所述输入图像的纹理分布特征与数据库中每一幅图像的 纹理分布特征的相似度，得到输入图像与数据库中每一幅图像之间的 纹理分布特征相似度 Sb (1)， 1取 0、 1、 2...数据库图像总数 -1; Calculating texture distribution characteristics of the input image and each image in the database separately The similarity of the texture distribution features, the texture distribution feature similarity between the input image and each image in the database Sb (1), 1 takes 0, 1, 2... the total number of database images -1;
歩骤 3： 利用公式 S (i) =WaxSa (i) +Wb Sb (i)， i取 0、 1、 2... 数据库图像总数 -1， Wa、 Wb为加权系数且 Wa+Wb=l， 计算输入图 像与数据库中每一幅图像的组合相似度 S (i)0 优选地， 所述颜色分布特征的获取方法包括： Step 3: Using the formula S (i) = WaxSa (i) + Wb Sb (i), i takes 0, 1, 2... The total number of database images -1, Wa, Wb are weighting coefficients and Wa + Wb = l calculating a combination of the input image and each image database similarity S (i) 0 preferably, the method of obtaining a color distribution comprising:
歩骤 201: 将图像转换到 颜色空间， 得到图像 I；  Step 201: Convert an image to a color space to obtain an image I;
歩骤 202:将图像各个像素的 H、 S, 分量映射为颜色特征值 G: Step 202: Mapping H, S, and components of each pixel of the image into color feature values G:
G = Qs*Qv*H + Qv*S + V, 将 颜色空间的三个通道的取值范围进行 区间划分， 分别划分为 A , 其中0≤ζ≤¾,0≤·/≤ ,0≤ ， ， ， 分别表示 颜色空间的三个通道被分割的区间总数； 歩骤 203: 统计图像中各个像素点的特征值分布情况： 遍历每个 像素点的颜色特征值，统计落入各个颜色分布直方图区间的像素点数 量，将落入各个颜色分布直方图区间的像素点数量分别除以图像像素 点总数,得到归一化的颜色分布特征^t(x)， 其中 X代表颜色分布直方 图区间。 优选地， 所述颜色分布特征的获取方法还包括： G = Q s *Q v *H + Q v *S + V, divide the range of values of the three channels of the color space into sections, respectively, into A, where 0≤ ζ ≤ 3⁄4, 0≤ ·/≤, 0 ≤ , , , respectively represent the total number of intervals in which the three channels of the color space are divided; Step 203: Statistic value distribution of each pixel in the statistical image: traverse the color feature value of each pixel, and the statistics fall into each color The number of pixels in the distribution histogram interval is divided by the number of pixels in each color distribution histogram interval divided by the total number of image pixels to obtain a normalized color distribution feature ^t(x), where X represents the color distribution histogram Diagram interval. Preferably, the method for acquiring the color distribution feature further includes:
将图像划分为 Ν块； 在所述歩骤 203中： 统计图像中各个像素点 的特征值分布情况： 遍历每个像素点的特征值， 统计落入各个颜色分 布直方图区间的像素点数量，且将不是图像边界块中的像素点统计两 次；将落入各个颜色分布直方图区间的像素点数量分别除以图像像素 点总数,得到归一化的颜色分布特征^t(x)， 其中 X代表颜色直方图。 优选地， 所述纹理分布特征的获取方法包括： 歩骤 301 : 将图像转换为灰度图， 得到图像 L; Dividing the image into Ν blocks; In the step 203: statistic value distribution of each pixel in the statistical image: traversing the eigenvalue of each pixel, and counting the number of pixels falling into the histogram interval of each color distribution, And the pixel points in the image boundary block are not counted twice; the number of pixels falling into the histogram interval of each color distribution is divided by the total number of image pixels, and a normalized color distribution feature ^t(x) is obtained, wherein X represents the color histogram. Preferably, the method for obtaining the texture distribution feature includes: Step 301: Convert an image into a grayscale image to obtain an image L;
歩骤 302: 以尺寸为 3像素 χ3像素的模板， 遍历所述图像 得 到每个模板的 LHP特征， 其中得到模板 LBp特征的方法包括： Step 302: traversing the image to obtain an LHP feature of each template by using a template with a size of 3 pixels and 3 pixels, wherein the method for obtaining the template L B p feature includes:
记模板中的 9个像素点的灰度值为 P' (Q≤ z≤ 8)， 其中模板正中的像 素灰度值记为 ； 将模板中其它的像素点的灰度值减去 。得到： The gray value of the nine pixels in the template is P' ( Q ≤ z ≤ 8 ), where the gray value of the pixel in the middle of the template is recorded as; the gray value of the other pixels in the template is subtracted. get:
对每个计算得到的 &进行二值化处理： 如果 g' ≥Q则令 g' = 1， 否则 g- = 0 ;将位于位置 z的像素的 &值扩展为 8位 2进制数，得到 特 征为， 1≤ ≤8： g 0 . For each calculated & binarization: If g' ≥ Q then let g' = 1 , otherwise g - = 0 ; expand the & value of the pixel at position z to an 8-bit binary number, The characteristic is, 1≤ ≤8: g 0 .
歩骤 303 : 获得每个模板的旋转不变的 特征； 其中得到模板 的旋转不变的 LBP"特征的方法包括：  Step 303: Obtaining a rotation-invariant feature of each template; wherein the method of obtaining the rotation-invariant LBP feature of the template includes:
对模板的每个 H(0按进行移位操作，分别可以得到 8个二进制数 据， 取其中最小的一个作为旋转不变的 LBP" 特征： For each H of the template (0 is shifted by operation, respectively, 8 binary data can be obtained, and the smallest one is taken as the rotation-invariant LBP ".
LBP" (/) = q)) , 式中 l≤z'≤8， RC>R表示移位操作， q
Figure imgf000005_0001
表示移位位数； LBP" (/) = q)) , where l ≤ z' ≤ 8, RC > R represents the shift operation, q
Figure imgf000005_0001
Indicates the number of shift bits;
歩骤 304 : 统计各个模板中各个旋转不变的^ 特征的分布情 况：遍历各个模板的每个旋转不变的 特征值，统计落入各个纹 理分布直方图区间的像素点数量，再将落入各个纹理分布直方图区间 的像素点数量分别除以图像像素点总数,得到归一化的纹理分布特征 hist(y) , 其中 _y代表纹理分布直方图区间。 优选地，所述歩骤 2中计算颜色分布特征相似度 Sa的方法包括: 歩骤 401 : 利用公式Step 304: Counting the distribution of each of the rotation-invariant features in each template: traversing each rotation-invariant feature value of each template, and counting the number of pixels falling into each texture distribution histogram interval, and then falling into The number of pixels in each texture distribution histogram interval is divided by the total number of image pixels, and the normalized texture distribution feature hist(y) is obtained, where _y represents the texture distribution histogram interval. Preferably, the method for calculating the color distribution feature similarity Sa in the step 2 includes: Step 401: Using the formula
Figure imgf000006_0001
相似度， 其中 tet x)为第一幅图像的颜色分布特征， tot2 (x)为第二幅 图像的颜色分布特征。 优选地，所述歩骤 2中计算纹理分布特征相似度 Sb的方法包括：
Figure imgf000006_0001
Similarity, where tet x) is the color distribution characteristic of the first image, and tot 2 (x) is the color distribution characteristic of the second image. Preferably, the method for calculating the texture distribution feature similarity Sb in the step 2 includes:
^ mrn^is^ (y\ hist2 y)) ^ mrn^is^ (y\ hist 2 y))
歩骤 401 : 利用公式& =^ 计算纹理分布特征相 似度， 其中 Ο 为第一幅图像的纹理分布特征， 为第二幅图 像的纹理分布特征。 优选地， 所述 Wa>Wb。 本发明还保护一种用于纹理分布图像检索方法的相似度获取方 法， 包括：  Step 401: Calculate the similarity of the texture distribution feature using the formula & =^, where Ο is the texture distribution feature of the first image and the texture distribution feature of the second image. Preferably, the Wa>Wb. The present invention also protects a similarity acquisition method for a texture distribution image retrieval method, including:
歩骤 1 : 提取输入图像的纹理分布特征；  Step 1: extracting a texture distribution feature of the input image;
歩骤 2: 分别计算所述输入图像的纹理分布特征与数据库中每一 幅图像的纹理分布特征的相似度，得到输入图像与数据库中每一幅图 像之间的纹理分布特征相似度 Sb ( i)， 1取 0、 1、 2...数据库图像总 数 -1 ; 所述纹理分布特征的获取方法包括：  Step 2: Calculate the similarity between the texture distribution feature of the input image and the texture distribution feature of each image in the database, and obtain the texture distribution feature similarity Sb between the input image and each image in the database (i) ), 1 takes 0, 1, 2... total number of database images -1; the method for obtaining the texture distribution features includes:
歩骤 301 : 将图像转换为灰度图， 得到图像 L;  Step 301: Converting an image into a grayscale image to obtain an image L;
歩骤 302: 以尺寸为 3像素 χ3像素的模板， 遍历所述图像 得 到每个模板的 LHP特征， 其中得到模板 LBp特征的方法包括： 记模板中的 9个像素点的灰度值为 P' (Q≤ z≤ 8)， 其中模板正中的像 素灰度值记为 ； 将模板中其它的像素点的灰度值减去 。得到： 对每个计算得到的 &进行二值化处理： 如果 g' ≥Q则令 g' = 1， 否则 g- = 0 ;将位于位置 z的像素的 &值扩展为 8位 2进制数，得到 特 征为， 1≤ ≤8： g 0 . Step 302: traversing the image to obtain an LHP feature of each template by using a template with a size of 3 pixels and 3 pixels, wherein the method for obtaining the template L B p feature includes: recording a gray value of 9 pixels in the template P' ( Q ≤ z ≤ 8 ), where the gray value of the pixel in the middle of the template is recorded as; the gray value of the other pixels in the template is subtracted. get: For each calculated & binarization: If g' ≥ Q then let g' = 1 , otherwise g - = 0 ; expand the & value of the pixel at position z to an 8-bit binary number, The characteristic is, 1≤ ≤8: g 0 .
歩骤 303 : 获得每个模板的旋转不变的 特征； 其中得到模板 的旋转不变的 LBP"特征的方法包括：  Step 303: Obtaining a rotation-invariant feature of each template; wherein the method of obtaining the rotation-invariant LBP feature of the template includes:
对模板的每个 H(0按进行移位操作，分别可以得到 8个二进制数 据， 取其中最小的一个作为旋转不变的 LBP" 特征： For each H of the template (0 is shifted by operation, respectively, 8 binary data can be obtained, and the smallest one is taken as the rotation-invariant LBP ".
LBP" (i) = m[n(ROR(LBP(il q)) , 式中 1≤ 8， ROR表示移位操作， g 0 表示移位位数； LBP" (i) = m [ n (ROR(LBP(il q)) , where 1 ≤ 8, ROR represents the shift operation, and g 0 represents the number of shift bits;
歩骤 304: 统计各个模板中各个旋转不变的^ 特征的分布情 况：遍历各个模板的每个旋转不变的 特征值，统计落入各个纹 理分布直方图区间的像素点数量，再将落入各个纹理分布直方图区间 的像素点数量分别除以图像像素点总数,得到归一化的纹理分布特征 hist(y) , 其中 _y代表纹理分布直方图区间。  Step 304: Counting the distribution of each of the rotation-invariant features in each template: traversing each rotation-invariant feature value of each template, and counting the number of pixels falling into each texture distribution histogram interval, and then falling into The number of pixels in each texture distribution histogram interval is divided by the total number of image pixels, and the normalized texture distribution feature hist(y) is obtained, where _y represents the texture distribution histogram interval.
本发明还保护上述中的纹理分布特征的获取方法。 综上所述， 由于采用了上述技术方案， 本发明的有益效果是： 本发明涉及的图像相似度获取方法不需对图像进行任何假设， 也 不需要大量的标注样本训练模型，具有容易实现，计算速度快的优点。 附图说明  The present invention also protects the method of acquiring texture distribution features in the above. In summary, the above technical solution is adopted, and the beneficial effects of the present invention are as follows: The image similarity obtaining method according to the present invention does not need to make any assumptions on the image, and does not require a large number of labeled sample training models, and is easy to implement. The advantage of fast calculation. DRAWINGS
本发明将通过例子并参照附图的方式说明， 其中： 图 1为本发明中颜色分布特征提取流程图。 The invention will be illustrated by way of example and with reference to the accompanying drawings in which: FIG. 1 is a flow chart of color distribution feature extraction in the present invention.
图 2为本发明中纹理分布特征提取流程图。 图 3为本发明中图像检索流程图。 具体实施方式 本说明书中公开的所有特征，或公开的所有方法或过程中的歩骤， 除了互相排斥的特征和 /或歩骤以外， 均可以以任何方式组合。 本说明书中公开的任一特征， 除非特别叙述， 均可被其他等效或 具有类似目的的替代特征加以替换。 gp， 除非特别叙述， 每个特征只 是一系列等效或类似特征中的一个例子而已。  2 is a flow chart of texture distribution feature extraction in the present invention. Figure 3 is a flow chart of image retrieval in the present invention. DETAILED DESCRIPTION OF THE INVENTION All of the features disclosed in this specification, or steps in all methods or processes disclosed, can be combined in any manner other than mutually exclusive features and/or steps. Any feature disclosed in this specification, unless specifically stated otherwise, may be replaced by other equivalent features or similar features. Gp, unless specifically stated, each feature is only one example of a series of equivalent or similar features.
本发明提供了一种用于颜色分布和纹理分布图像检索的相似度获 取方法， 其具体歩骤包括：  The present invention provides a similarity acquisition method for color distribution and texture distribution image retrieval, and the specific steps include:
歩骤 1: 提取输入图像的颜色分布特征及纹理分布特征； 歩骤 2: 分别计算所述输入图像的颜色分布特征与数据库中每一 幅图像的颜色分布特征的相似度，得到输入图像与数据库中每一幅图 像之间的颜色分布特征相似度 Sa (0, i取 0、 1、 2...数据库图像总 数 -1;  Step 1: extracting the color distribution feature and the texture distribution feature of the input image; Step 2: respectively calculating the similarity between the color distribution feature of the input image and the color distribution feature of each image in the database, and obtaining an input image and a database The color distribution characteristic similarity Sa between each image in each case (0, i takes 0, 1, 2... total number of database images -1;
分别计算所述输入图像的纹理分布特征与数据库中每一幅图像的 纹理分布特征的相似度，得到输入图像与数据库中每一幅图像之间的 纹理分布特征相似度 Sb (i)， 1取 0、 1、 2...数据库图像总数 -1; 歩骤 3: 利用公式 S (i) =WaxSa (i) +WbxSb (i)， i取 0、 1、 2...数据库图像总数-1， Wa、 Wb为加权系数且 Wa+Wb=l， 计算输入 图像与数据库中每一幅图像的组合相似度 S (i)0 由于人们在一般情 况下更关心颜色， 因此作为一个优选的实施方式， 加权系数 Wa>Wb。 如图 3， 当获得了输入图像与数据库每一幅图像之间的组合相似 度后， 对各个相似度进行排序， 相似度越大说明两幅图像越相似， 我 们可以根据经验设定一阈值，将大于组合相似度大于该阈值的所有数 据库中的图像输出， 作为检索结果。 如图 1，在本发明一个实施例中，颜色分布特征的获取方法包括： 歩骤 201: 将图像转换到 颜色空间， 得到图像 I； 一般来说图 片为 RGB颜色空间，将 RGB颜色空间的图片转换到 颜色空间为 本领域公知的技术， 在此不再赘述其具体过程。 Calculating the similarity between the texture distribution feature of the input image and the texture distribution feature of each image in the database, and obtaining the texture distribution feature similarity Sb (i), 1 between the input image and each image in the database. 0, 1, 2... Total number of database images -1; Step 3: Using the formula S (i) = WaxSa (i) + WbxSb (i), i takes 0, 1, 2... Total number of database images -1 , Wa, Wb and the weighting coefficient Wa + Wb = l, calculate the combination of the input image and each image database similarity S (i) 0 as people are more concerned with color in general, and therefore as a preferred embodiment , weighting coefficient Wa>Wb. As shown in Figure 3, when the combined similarity between the input image and each image of the database is obtained, each similarity is sorted. The greater the similarity, the more similar the two images are, and we can set a threshold based on experience. An image output in all databases larger than the combined similarity is greater than the threshold as a retrieval result. As shown in FIG. 1, in an embodiment of the present invention, a method for acquiring a color distribution feature includes: Step 201: Convert an image into a color space to obtain an image I; generally, the image is an RGB color space, and the image of the RGB color space is obtained. The conversion to the color space is a technique well known in the art, and the specific process thereof will not be described herein.
歩骤 202 : 将图像各个像素的 H、 S , 分量按照公式 G = * *H + 2V^+ 映射关系转换为颜色特征值 G; 其中， Qh,Qs,Qv 的定义是这样的：将^^颜色空间的三个通道的取值范围进行区间划 分， 分别划分为 ， Λ， 其中 o≤ ≤ ,o≤_/≤ ,o≤A≤a， , a, 分 别表示 HSV颜色空间的三个通道被分割的区间总数； 歩骤 203: 统计图像中各个像素点的特征值分布情况： 遍历每个 像素点的颜色特征值，统计落入各个颜色分布直方图区间的像素点数 量，将落入各个颜色分布直方图区间的像素点数量分别除以图像像素 点总数,得到归一化的颜色分布特征^t(x)， 其中 X代表颜色分布直方 图区间。 本领域技术人员均知晓，颜色分布直方图将整幅图像的颜色特征 分为若干区间，然后用各个像素在各个区间分布的情况描述不同色彩 在整幅图像中所占的比例。 考虑到一幅图像所表达的含义，往往位于图像边界附近的区域不 太重要， 我们更关心图非边界区域表达的内容。 因此， 在本发明另一 个实施例中， 所述颜色分布特征的获取方法还包括： Step 202: Convert H, S, and components of each pixel of the image into a color feature value G according to a formula G = * * H + 2 V ^ + ; wherein Q h , Q s , Q v are defined as follows : The range of values of the three channels of the ^^ color space is divided into sections, respectively, where ≤, where o ≤ ≤ , o ≤ _ / ≤ , o ≤ A ≤ a, , a, respectively represent the HSV color space The total number of intervals in which the three channels are divided; Step 203: Statistic value distribution of each pixel in the statistical image: traverse the color feature values of each pixel, and count the number of pixels falling into the histogram interval of each color distribution, The number of pixels falling into the histogram interval of each color distribution is divided by the total number of image pixels, respectively, to obtain a normalized color distribution feature ^t(x), where X represents a color distribution histogram interval. Those skilled in the art are aware that the color distribution histogram divides the color features of the entire image into several intervals, and then describes the proportion of different colors in the entire image by the case where each pixel is distributed in each interval. Considering the meaning expressed by an image, the area that is often near the boundary of the image is not Too important, we are more concerned with the content expressed in the non-boundary areas of the map. Therefore, in another embodiment of the present invention, the method for acquiring the color distribution feature further includes:
将图像划分为 N块， 例如 N等于 36; 包含在图像边界块中的像 素仅被统计一次， 其余像素被统计两次。 具体而言， 在所述歩骤 203 中： 统计图像中各个像素点的特征值分布情况： 遍历每个像素点的特 征值， 统计落入各个颜色分布直方图区间的像素点数量； 且将不是图 像边界块中的像素点统计两次， gp， 当有不是图像边界块中的像素点 的特征值落入某一颜色分布直方图区间时，将落入该区间的像素点数 量加 2; 当有是图像边界块中的像素点的特征值落入某一颜色分布直 方图区间时， 则将落入该区间的像素点数量加 1 ; 最后再将落入各个 颜色分布直方图区间的像素点数量分别除以图像像素点总数,得到归 一化的颜色分布特征 tet(x)， 其中 X代表颜色直方图。  The image is divided into N blocks, for example, N is equal to 36; the pixels contained in the image boundary block are counted only once, and the remaining pixels are counted twice. Specifically, in the step 203: collecting the feature value distribution of each pixel in the image: traversing the feature value of each pixel, and counting the number of pixels falling into the histogram interval of each color distribution; The pixel points in the image boundary block are counted twice, gp. When there is a feature value of a pixel in the image boundary block that falls within a certain color distribution histogram interval, the number of pixels falling into the interval is increased by 2; When the feature value of the pixel in the image boundary block falls within a certain color distribution histogram interval, the number of pixels falling into the interval is increased by 1; finally, the pixel falling into the histogram interval of each color distribution is added. The number is divided by the total number of image pixels to obtain a normalized color distribution characteristic tet(x), where X represents a color histogram.
这样统计出来的颜色分布特征值更加准确。 如图 2， 在本发明的另一个实施例中， 所述纹理分布特征的获取 方法包括：  The color distribution feature values thus calculated are more accurate. As shown in FIG. 2, in another embodiment of the present invention, the method for obtaining the texture distribution feature includes:
歩骤 301 : 将图像转换为灰度图， 得到图像 L; 将 RGB图像转换 为 灰 度 图 有 多 种 方 法 ， 其 中 一 种 是 利 用 公 式 Ζ = 0.299^ + 0.587 *(? + 0.114 进行转换， 其中， R代表像素的红色分 量， G代表像素的绿色分量， Β代表像素的蓝色分量。 0.299、 0.587、 0.114为系数， 当然这个系数并不唯一， 不能理解为对本发明的限制。 Step 301: Converting an image into a grayscale image to obtain an image L; There are various methods for converting an RGB image into a grayscale image, one of which is converted by using the formula Ζ = 0.299^ + 0.587 *(? + 0.114 R represents the red component of the pixel, G represents the green component of the pixel, and Β represents the blue component of the pixel. 0.299, 0.587, 0.114 are coefficients, although this coefficient is not unique and cannot be construed as limiting the invention.
歩骤 302: 以尺寸为 3像素 χ3像素的模板， 遍历所述图像 得 到每个模板的^^特征 （即纹理特征）， 其中得到模板^^特征的方 法包括： 记模板中的 9个像素点的灰度值为 P'(Q≤ z≤ 8)， 其中模板正中的像 素灰度值记为 ； 将模板中其它的像素点的灰度值减去 。得到： Step 302: traversing the image to obtain a ^^ feature (ie, a texture feature) of each template by using a template with a size of 3 pixels and 3 pixels, wherein the method for obtaining the template feature includes: recording 9 pixels in the template The gray value is P' ( Q ≤ z ≤ 8 ), where the image in the middle of the template The prime gray value is recorded as; the gray value of other pixels in the template is subtracted. get:
对每个计算得到的 &进行二值化处理： 如果 g' ≥Q则令 g' = 1， 否则 g- = 0 ;将位于位置 z的像素的 &值扩展为 8位 2进制数，得到 特 征为， 1≤ ≤8：
Figure imgf000011_0001
For each calculated & binarization: If g' ≥ Q then let g' = 1 , otherwise g - = 0 ; expand the & value of the pixel at position z to an 8-bit binary number, The characteristic is, 1≤ ≤8:
Figure imgf000011_0001
上述计算的^ ^特征不能应对旋转不变的要求， 为了获得旋转不 变的 Z ?P特征， 需要进一歩执行歩骤 303 : 获得每个模板的旋转不变 的 LBP"特征； 其中得到模板的旋转不变的 LBP"特征的方法包括： 对模板的每个 H(0按进行移位操作，分别可以得到 8个二进制数 据， 取其中最小的一个作为旋转不变的 LBP" 特征： The ^^ feature of the above calculation cannot cope with the requirement of rotation invariance. In order to obtain the rotation-invariant Z?P feature, it is necessary to perform step 303: obtaining the rotation-invariant LBP" feature of each template; The method of rotating the invariant LBP "features includes: For each H of the template (0 is shifted by operation, respectively, 8 binary data can be obtained, and the smallest one is taken as the rotation-invariant LBP ".
LBP" (i) = m[n(ROR(LBP(il q)) , 式中 1≤ 8， ROR表示移位操作， g 0 表示移位位数； LBP" (i) = m [ n (ROR(LBP(il q)) , where 1 ≤ 8, ROR represents the shift operation, and g 0 represents the number of shift bits;
歩骤 304: 统计各个模板中各个旋转不变的^ 特征的分布情 况：遍历各个模板的每个旋转不变的 特征值，统计落入各个纹 理分布直方图区间的像素点数量，再将落入各个纹理分布直方图区间 的像素点数量分别除以图像像素点总数,得到归一化的纹理分布特征 hist(y) , 其中 _y代表纹理分布直方图区间。 这里纹理分布直方图的定义与前述颜色分布直方图定义类似。纹 理分布直方图将整幅图像的纹理特征分为若干区间，然后用各个像素 在各个区间分布的情况描述不同纹理在整幅图像中所占的比例。  Step 304: Counting the distribution of each of the rotation-invariant features in each template: traversing each rotation-invariant feature value of each template, and counting the number of pixels falling into each texture distribution histogram interval, and then falling into The number of pixels in each texture distribution histogram interval is divided by the total number of image pixels, and the normalized texture distribution feature hist(y) is obtained, where _y represents the texture distribution histogram interval. The definition of the texture distribution histogram here is similar to the definition of the aforementioned color distribution histogram. The texture distribution histogram divides the texture features of the entire image into several intervals, and then describes the proportion of different textures in the entire image by the case where each pixel is distributed in each interval.
当得到图像的颜色分布特征后，计算两幅图像的颜色分布特征相 似度 Sa的一个具体实施方式， 包括： 歩骤 401 : 利用公式After obtaining the color distribution characteristics of the image, calculating the color distribution characteristics of the two images A specific implementation of the degree of Sa, including: Step 401: Using the formula
Figure imgf000012_0001
相似度， 其中 tet x)为第一幅图像的颜色分布特征， tot2 (x)为第二幅 图像的颜色分布特征， 其中 X代表颜色直方图。 当得到图像的纹理分布特征后，计算两幅图像的纹理分布特征相 似度 Sb的一个具体实施方式， 包括：
Figure imgf000012_0001
Similarity, where tet x) is the color distribution characteristic of the first image, and tot 2 (x) is the color distribution characteristic of the second image, where X represents the color histogram. After obtaining the texture distribution feature of the image, a specific implementation manner of calculating the texture distribution feature similarity Sb of the two images includes:
^ mrn^is^ (y\ hist2 y)) ^ mrn^is^ (y\ hist 2 y))
歩骤 501 : 利用公式& =^ 计算纹理分布特征相 似度， 其中 Ο 为第一幅图像的纹理分布特征， 0 为第二幅图 像的纹理分布特征。 J代表纹理分布直方图区间。 在上述内容的教导下，本领域技术人员容易想到基于本发明创新 思想的一种用于纹理分布图像检索方法的相似度获取方法， 包括： 歩骤 1 : 提取输入图像的纹理分布特征；  Step 501: Calculate the similarity of the texture distribution feature using the formula & =^, where Ο is the texture distribution feature of the first image, and 0 is the texture distribution feature of the second image. J represents the texture distribution histogram interval. Under the above teachings, a person skilled in the art can easily think of a similarity acquisition method for a texture distribution image retrieval method based on the innovative idea of the present invention, comprising: Step 1: Extracting a texture distribution feature of an input image;
歩骤 2 : 分别计算所述输入图像的纹理分布特征与数据库中每一 幅图像的纹理分布特征的相似度，得到输入图像与数据库中每一幅图 像之间的纹理分布特征相似度 Sb ( i)， 1取 0、 1、 2. . .数据库图像总 数 -1。 同理，当获得了输入图像与数据库每一幅图像之间的纹理分布特 征相似度后， 对各个相似度进行排序， 相似度越大说明两幅图像越相 似， 我们可以根据经验设定一阈值， 将大于组合相似度大于该阈值的 所有数据库中的图像输出， 作为检索结果。 本发明并不局限于前述的具体实施方式。 本发明扩展到任何在本 说明书中披露的新特征或任何新的组合，以及披露的任一新的方法或 过程的歩骤或任何新的组合。 Step 2: respectively calculating the similarity between the texture distribution feature of the input image and the texture distribution feature of each image in the database, and obtaining a texture distribution feature similarity Sb between the input image and each image in the database (i) ), 1 takes 0, 1, 2. 2. The total number of database images -1. Similarly, when the texture distribution feature similarity between the input image and each image of the database is obtained, each similarity is sorted. The greater the similarity, the more similar the two images are. We can set a threshold according to experience. , will output the image in all databases larger than the combined similarity greater than the threshold as the search result. The invention is not limited to the specific embodiments described above. The invention extends to any novel features or any new combinations disclosed in this specification, as well as a novel or a novel combination of any new method or process disclosed.

Claims

权 利 要 求 书 claims
1、 一种用于颜色分布和纹理分布图像检索的相似度获取方法，其 特征在于， 包括： 1. A similarity acquisition method for color distribution and texture distribution image retrieval, which is characterized by: including:
歩骤 1: 提取输入图像的颜色分布特征及纹理分布特征； 歩骤 2: 分别计算所述输入图像的颜色分布特征与数据库中每一 幅图像的颜色分布特征的相似度，得到输入图像与数据库中每一幅图 像之间的颜色分布特征相似度 Sa (0, i取 0、 1、 2...数据库图像总 数 -1; Step 1: Extract the color distribution characteristics and texture distribution characteristics of the input image; Step 2: Calculate the similarity between the color distribution characteristics of the input image and the color distribution characteristics of each image in the database to obtain the input image and the database The color distribution feature similarity between each image in Sa (0, i is 0, 1, 2... The total number of database images - 1;
分别计算所述输入图像的纹理分布特征与数据库中每一幅图像的 纹理分布特征的相似度，得到输入图像与数据库中每一幅图像之间的 纹理分布特征相似度 Sb (i)， 1取 0、 1、 2...数据库图像总数 -1; Calculate the similarity between the texture distribution features of the input image and the texture distribution features of each image in the database respectively to obtain the texture distribution feature similarity Sb (i) between the input image and each image in the database, 1 is taken 0, 1, 2...total number of database images -1;
歩骤 3： 利用公式 S (i) =WaxSa (i) +Wb Sb (i)， i取 0、 1、 2... 数据库图像总数 -1， Wa、 Wb为加权系数且 Wa+Wb=l， 计算输入图 像与数据库中每一幅图像的组合相似度 S (i)0 Step 3: Use the formula S (i) =WaxSa (i) +Wb Sb (i), i is 0, 1, 2... The total number of database images is -1, Wa and Wb are weighting coefficients and Wa+Wb=l , calculate the combined similarity S (i) 0 between the input image and each image in the database
2、 根据权利要求 1 所述的用于颜色分布和纹理分布图像检索的 相似度获取方法， 其特征在于， 所述颜色分布特征的获取方法包括： 歩骤 201: 将图像转换到 颜色空间， 得到图像 I； 2. The similarity acquisition method for color distribution and texture distribution image retrieval according to claim 1, characterized in that the acquisition method of the color distribution characteristics includes: Step 201: Convert the image to color space to obtain imageI;
歩骤 202:将图像各个像素的 H、 S, 分量映射为颜色特征值 G: G = Qs*Qv*H + Qv*S + V, 将 颜色空间的三个通道的取值范围进行 区间划分， 分别划分为 ΑΛ， 其中0≤ζ≤¾,0≤·/≤ ,0≤ ， ， ， 分别表示 颜色空间的三个通道被分割的区间总数； 歩骤 203: 统计图像中各个像素点的颜色特征值分布情况： 遍历 每个像素点的颜色特征值，统计落入各个颜色分布直方图区间的像素 点数量，将落入各个颜色分布直方图区间的像素点数量分别除以图像 像素点总数,得到归一化的颜色分布特征^t(x)， 其中 X代表颜色分布 直方图区间。 Step 202: Map the H, S, components of each pixel of the image to the color feature value G: G = Q s *Q v *H + Q v *S + V, and carry out the value range of the three channels of the color space. Interval division is divided into Α Λ respectively, where 0≤ ζ ≤ ¾, 0≤ ·/≤ ,0≤ , , , respectively represent the total number of intervals divided by the three channels of the color space; Step 203: Count each pixel in the image Distribution of color feature values of points: Traverse the color feature values of each pixel, count the number of pixels falling into each color distribution histogram interval, and divide the number of pixels falling into each color distribution histogram interval by the image pixels The total number of points is used to obtain the normalized color distribution feature ^t(x), where X represents the color distribution histogram interval.
3、 根据权利要求 2 所述的用于颜色分布和纹理分布图像检索的 相似度获取方法，其特征在于，所述颜色分布特征的获取方法还包括: 将图像划分为 N块； 在所述歩骤 203中： 统计图像中各个像素点 的特征值分布情况： 遍历每个像素点的特征值， 统计落入各个颜色分 布直方图区间的像素点数量，且将不是图像边界块中的像素点统计两 次；将落入各个颜色分布直方图区间的像素点数量分别除以图像像素 点总数,得到归一化的颜色分布特征^t(x)， 其中 X代表颜色直方图。 3. The similarity acquisition method for color distribution and texture distribution image retrieval according to claim 2, characterized in that the acquisition method of color distribution characteristics further includes: dividing the image into N blocks; in the step Step 203: Count the distribution of eigenvalues of each pixel in the image: Traverse the eigenvalues of each pixel and count the number of pixels falling into each color distribution histogram interval, which will not be the statistics of pixels in the image boundary block. Twice; divide the number of pixels falling into each color distribution histogram interval by the total number of image pixels to obtain the normalized color distribution feature ^t(x), where X represents the color histogram.
4、 根据权利要求 1或 2所述的用于颜色分布和纹理分布图像检索 的相似度获取方法，其特征在于，所述纹理分布特征的获取方法包括: 歩骤 301 : 将图像转换为灰度图， 得到图像 L; 4. The similarity acquisition method for color distribution and texture distribution image retrieval according to claim 1 or 2, characterized in that the acquisition method of texture distribution features includes: Step 301: Convert the image to grayscale Figure, get image L;
歩骤 302: 以尺寸为 3像素 χ3像素的模板， 遍历所述图像 得 到每个模板的 LHP特征， 其中得到模板 LBp特征的方法包括： Step 302: Using a template with a size of 3 pixels×3 pixels, traverse the image to obtain the LHP features of each template, where the method for obtaining the template L B p features includes:
记模板中的 9个像素点的灰度值为 P'(Q≤ z≤ 8)， 其中模板正中的像 素灰度值记为 ； 将模板中其它的像素点的灰度值减去 。得到： Record the gray value of the 9 pixels in the template as P' ( Q≤ z≤ 8 ), where the gray value of the pixel in the middle of the template is recorded as; Subtract the gray value of other pixels in the template. get:
对每个计算得到的 &进行二值化处理： 如果 g' ≥Q则令 g' = 1， 否则 g- = 0 ;将位于位置 z的像素的 &值扩展为 8位 2进制数，得到 特 征为， 1≤ ≤8：
Figure imgf000015_0001
Binarize each calculated &: If g' ≥ Q , then let g' = 1 , otherwise g - = 0 ; expand the & value of the pixel at position z into an 8-bit binary number, and get The characteristics are, 1≤ ≤8:
Figure imgf000015_0001
歩骤 303 : 获得每个模板的旋转不变的 特征； 其中得到模板 的旋转不变的 LBP"特征的方法包括： Step 303: Obtain the rotation-invariant features of each template; methods for obtaining the rotation-invariant LBP features of the template include:
对模板的每个 H(0按进行移位操作，分别可以得到 8个二进制数 据， 取其中最小的一个作为旋转不变的 LBPD特征： LBPn {i) = m[n(ROR(LBP(i), q)) , 式中 l≤z'≤8 ， ROR表示移位操作， g二 0 表示移位位数； Perform a shift operation on each H(0 of the template to obtain 8 binary data respectively, and take the smallest one as the rotation-invariant LBP D feature: LBP n {i) = m [ n (ROR(LBP(i), q)) , where l≤z'≤8, ROR represents the shift operation, and g＝0 represents the number of shifts;
歩骤 304: 统计各个模板中各个旋转不变的^ 特征的分布情 况：遍历各个模板的每个旋转不变的 特征值，统计落入各个纹 理分布直方图区间的像素点数量，再将落入各个纹理分布直方图区间 的像素点数量分别除以图像像素点总数,得到归一化的纹理分布特征 hist(y) , 其中 _y代表纹理分布直方图区间。 Step 304: Count the distribution of each rotation-invariant feature in each template: traverse each rotation-invariant feature value of each template, count the number of pixels that fall into each texture distribution histogram interval, and then count the number of pixels that fall into each texture distribution histogram interval. The number of pixels in each texture distribution histogram interval is divided by the total number of image pixels to obtain the normalized texture distribution feature hist(y), where _y represents the texture distribution histogram interval.
5、 根据权利要求 4 所述的用于颜色分布和纹理分布图像检索的 相似度获取方法， 其特征在于， 所述歩骤 2中计算颜色分布特征相似 度 Sa的方法包括： 歩骤 401 : 利用公式 5. The similarity acquisition method for color distribution and texture distribution image retrieval according to claim 4, characterized in that the method for calculating the color distribution feature similarity Sa in step 2 includes: Step 401: using formula
Figure imgf000016_0001
似度， 其中 ^ χ)为第一幅图像的颜色分布特征， 为第二幅图 像的颜色分布特征。
Figure imgf000016_0001
similarity, where ^ χ) is the color distribution characteristic of the first image, and is the color distribution characteristic of the second image.
6、 根据权利要求 4 所述的用于颜色分布和纹理分布图像检索的 相似度获取方法， 其特征在于， 所述歩骤 2中计算纹理分布特征相似 度 Sb的方法包括： 6. The similarity acquisition method for color distribution and texture distribution image retrieval according to claim 4, characterized in that the method for calculating the texture distribution feature similarity Sb in step 2 includes:
^ mrn^is^ (y\ hist2 y)) ^ mrn^is^ (y\ hist 2 y))
歩骤 401 : 利用公式& =^ ~~ ^-— 计算纹理分布特征相 似度， 其中 Ο 为第一幅图像的纹理分布特征， 0 为第二幅图 像的纹理分布特征。 Step 401: Use the formula & =^ ~~ ^-- to calculate the texture distribution feature similarity, where Ο is the texture distribution feature of the first image, and 0 is the texture distribution feature of the second image.
7、 根据权利要求 1 所述的用于颜色分布和纹理分布图像检索的 相似度获取方法， 其特征在于， 所述 Wa>Wb。 7. The similarity acquisition method for color distribution and texture distribution image retrieval according to claim 1, characterized in that, the Wa>Wb.
8、 一种用于纹理分布图像检索方法的相似度获取方法，其特征在 于， 包括： 8. A similarity acquisition method for texture distribution image retrieval method, which is characterized by including:
歩骤 1: 提取输入图像的纹理分布特征； Step 1: Extract texture distribution features of the input image;
歩骤 2: 分别计算所述输入图像的纹理分布特征与数据库中每一 幅图像的纹理分布特征的相似度， 到的若干纹理分布特征相似度 Sb (i)， 1取 0、 1、 2...数据库图像总数 -1; Step 2: Calculate the similarity between the texture distribution characteristics of the input image and the texture distribution characteristics of each image in the database respectively, and obtain the similarity Sb (i) of several texture distribution characteristics, 1 is 0, 1, 2. ..total number of database images -1;
所述纹理分布特征的获取方法包括： The method of obtaining the texture distribution characteristics includes:
歩骤 301: 将图像转换为灰度图， 得到图像 L; Step 301: Convert the image to grayscale to obtain image L;
歩骤 302: 以尺寸为 3像素 χ3像素的模板， 遍历所述图像 得 到每个模板的 LHP特征， 其中得到模板 LBP特征的方法包括： Step 302: Using a template with a size of 3 pixels×3 pixels, traverse the image to obtain the LHP features of each template, where the method of obtaining the template LBP features includes:
记模板中的 9个像素点的灰度值为 P'(Q≤ z≤ 8)， 其中模板正中的像 素灰度值记为 ； 将模板中其它的像素点的灰度值减去 。得到： Record the gray value of the 9 pixels in the template as P' ( Q≤ z≤ 8 ), where the gray value of the pixel in the middle of the template is recorded as; Subtract the gray value of other pixels in the template. get:
对每个计算得到的 &进行二值化处理： 如果 g'≥Q则令 g' =1， 否则 g-=0;将位于位置 z的像素的 &值扩展为 8位 2进制数，得到 特 征为， 1≤ ≤8：
Figure imgf000017_0001
Binarize each calculated &: If g' ≥ Q, let g' =1 , otherwise g - =0 ; expand the & value of the pixel at position z into an 8-bit binary number, and get The characteristics are, 1≤ ≤8:
Figure imgf000017_0001
g二 0 . g20.
歩骤 303: 获得每个模板的旋转不变的 特征； 其中得到模板 的旋转不变的 LBP"特征的方法包括： Step 303: Obtain the rotation-invariant features of each template; methods for obtaining the rotation-invariant LBP features of the template include:
对模板的每个 H(0按进行移位操作，分别可以得到 8个二进制数 据， 取其中最小的一个作为旋转不变的 LBP" 特征： Perform a shift operation on each H(0 of the template to obtain 8 binary data respectively, and take the smallest one as the rotation-invariant LBP " feature:
LBP" (/) = m[n(ROR(LBP(i), q)) , 式中 1≤ 8， RC>R表示移位操作， 表示移位位数； LBP" (/) = m [ n (ROR(LBP(i), q)) , where 1≤ 8, RC>R represents the shift operation, Represents the number of shifts;
歩骤 304: 统计各个模板中各个旋转不变的^ 特征的分布情 况：遍历各个模板的每个旋转不变的 特征值，统计落入各个纹 理分布直方图区间的像素点数量，再将落入各个纹理分布直方图区间 的像素点数量分别除以图像像素点总数,得到归一化的纹理分布特征 hist(y) , 其中 _y代表纹理分布直方图区间。 Step 304: Count the distribution of each rotation-invariant feature in each template: traverse each rotation-invariant feature value of each template, count the number of pixels that fall into each texture distribution histogram interval, and then count the number of pixels that fall into each texture distribution histogram interval. The number of pixels in each texture distribution histogram interval is divided by the total number of image pixels to obtain the normalized texture distribution feature hist(y), where _y represents the texture distribution histogram interval.
9、 根据权利要求 8 所述的用于纹理分布图像检索的相似度获取 方法， 其特征在于， 所述歩骤 2中计算纹理分布特征相似度 Sb的方 法包括： 9. The similarity acquisition method for texture distribution image retrieval according to claim 8, wherein the method for calculating the texture distribution feature similarity Sb in step 2 includes:
^ mrn^is^ (y\ hist2 y)) ^ mrn^is^ (y\ hist 2 y))
歩骤 401: 利用公式& =^ ~~ ^-— 计算纹理分布特征相 似度， 其中 Ο 为第一幅图像的纹理分布特征， 0 为第二幅图 像的纹理分布特征。 Step 401: Use the formula & =^ ~~ ^-- to calculate the texture distribution feature similarity, where Ο is the texture distribution feature of the first image, and 0 is the texture distribution feature of the second image.
10、 一种纹理分布特征的获取方法， 其特征在于， 包括： 歩骤 301: 将图像转换为灰度图， 得到图像 L; 10. A method for obtaining texture distribution characteristics, which is characterized by including: Step 301: Convert the image into a grayscale image to obtain image L;
歩骤 302: 以尺寸为 3像素 χ3像素的模板， 遍历所述图像 得 到每个模板的 LHP特征， 其中得到模板 LBp特征的方法包括： Step 302: Using a template with a size of 3 pixels × 3 pixels, traverse the image to obtain the LHP features of each template, where the method for obtaining the template L B p features includes:
记模板中的 9个像素点的灰度值为 P'(Q≤ Z≤ 8)， 其中模板正中的像 素灰度值记为 ； 将模板中其它的像素点的灰度值减去 。得到： Record the gray value of the 9 pixels in the template as P' ( Q≤ Z≤ 8 ), where the gray value of the pixel in the middle of the template is recorded as; Subtract the gray values of other pixels in the template. get:
对每个计算得到的 &进行二值化处理： 如果 g'≥ Q则令 g' = 1， 否则 g-=0;将位于位置 z的像素的 &值扩展为 8位 2进制数，得到 特 征为， 1≤ ≤8： g 0 . Binarize each calculated &: If g' ≥ Q , let g' = 1 , otherwise g - =0 ; expand the & value of the pixel at position z into an 8-bit binary number, and get The characteristics are, 1≤ ≤8: g 0 .
歩骤 303 : 获得每个模板的旋转不变的 特征； 其中得到模板 的旋转不变的 LBP"特征的方法包括： Step 303: Obtain the rotation-invariant features of each template; methods for obtaining the rotation-invariant LBP features of the template include:
对模板的每个 H(0按进行移位操作，分别可以得到 8个二进制数 据， 取其中最小的一个作为旋转不变的 LBP" 特征： Perform a shift operation on each H(0 of the template to obtain 8 binary data respectively, and take the smallest one as the rotation-invariant LBP " feature:
LBP" (i) = m[n(ROR(LBP(il q)) , 式中 1≤ 8， ROR表示移位操作， g 0 表示移位位数； LBP" (i) = m [ n (ROR(LBP(il q)) , where 1≤ 8, ROR represents the shift operation, g 0 represents the number of shift bits;
歩骤 304: 统计各个模板中各个旋转不变的^ 特征的分布情 况：遍历各个模板的每个旋转不变的 特征值，统计落入各个纹 理分布直方图区间的像素点数量，再将落入各个纹理分布直方图区间 的像素点数量分别除以图像像素点总数,得到归一化的纹理分布特征 hist(y) , 其中 _y代表纹理分布直方图区间。 Step 304: Count the distribution of each rotation-invariant feature in each template: traverse each rotation-invariant feature value of each template, count the number of pixels that fall into each texture distribution histogram interval, and then count the number of pixels that fall into each texture distribution histogram interval. The number of pixels in each texture distribution histogram interval is divided by the total number of image pixels to obtain the normalized texture distribution feature hist(y), where _y represents the texture distribution histogram interval.
PCT/CN2014/073995 2013-08-19 2014-03-25 Similarity acquisition method for colour distribution and texture distribution image retrieval WO2015024383A1 (en)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN2013103616152 2013-08-19
CN201310361615.2A CN103440646B (en) 2013-08-19 2013-08-19 Similarity acquisition methods for distribution of color and grain distribution image retrieval

Publications (1)

Publication Number Publication Date
WO2015024383A1 true WO2015024383A1 (en) 2015-02-26

Family

ID=49694339

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2014/073995 WO2015024383A1 (en) 2013-08-19 2014-03-25 Similarity acquisition method for colour distribution and texture distribution image retrieval

Country Status (2)

Country Link
CN (1) CN103440646B (en)
WO (1) WO2015024383A1 (en)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111461147A (en) * 2020-04-30 2020-07-28 柳州智视科技有限公司 Binary coding organization algorithm based on image characteristics
CN111861981A (en) * 2020-05-29 2020-10-30 北京工商大学 Image color space fat point detection and evaluation method based on point, line and surface
CN111985559A (en) * 2020-08-19 2020-11-24 合肥工业大学 Tire pattern structure similarity detection method based on boundary characteristics
CN114445659A (en) * 2022-01-20 2022-05-06 南京中医药大学 Method for identifying authenticity of spina date seeds based on image color and texture characteristics

Families Citing this family (16)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103440646B (en) * 2013-08-19 2016-08-10 成都品果科技有限公司 Similarity acquisition methods for distribution of color and grain distribution image retrieval
CN104699726B (en) * 2013-12-18 2018-03-23 杭州海康威视数字技术股份有限公司 A kind of vehicle image search method and device applied to traffic block port
CN104750697B (en) * 2013-12-27 2019-01-25 同方威视技术股份有限公司 Searching system, search method and Security Inspection Equipments based on fluoroscopy images content
CN104156732A (en) * 2014-08-01 2014-11-19 北京利云技术开发公司 Paper authenticity identification system and method
CN105574837A (en) * 2014-10-13 2016-05-11 Tcl集团股份有限公司 Image similarity matching method and device
CN106033549B (en) * 2015-03-16 2019-05-07 北京大学 Method for reordering and device in a kind of vehicle retrieval
CN105160029B (en) * 2015-09-30 2019-04-16 努比亚技术有限公司 Image retrieving apparatus and method
CN107145487A (en) * 2016-03-01 2017-09-08 深圳中兴力维技术有限公司 Image search method and device
CN106202250B (en) * 2016-06-29 2019-12-17 西安电子科技大学 Image retrieval method based on class comprehensive characteristics and complete class matching
CN107729855B (en) * 2017-10-25 2022-03-18 成都尽知致远科技有限公司 Mass data processing method
CN108876832B (en) * 2018-05-30 2022-04-26 重庆邮电大学 Robust texture feature extraction method based on grouping-order mode
CN109671113B (en) * 2018-09-29 2022-12-20 浙江华孚色纺有限公司 Color measuring and matching method of colored spun fabric based on image technology
CN110298366B (en) * 2019-07-05 2021-05-04 北华航天工业学院 Crop distribution extraction method and device
CN110909791B (en) * 2019-11-20 2023-02-28 车智互联（北京）科技有限公司 Similar image identification method and computing device
CN115830352B (en) * 2023-02-20 2023-06-02 深圳中微电科技有限公司 Image similarity comparison method, device and storage medium
CN116403419B (en) * 2023-06-07 2023-08-25 贵州鹰驾交通科技有限公司 Traffic light control method based on vehicle-road cooperation

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20080166026A1 (en) * 2007-01-10 2008-07-10 Samsung Electronics Co., Ltd. Method and apparatus for generating face descriptor using extended local binary patterns, and method and apparatus for face recognition using extended local binary patterns
CN101425136A (en) * 2008-10-24 2009-05-06 北京工业大学 Method and device for acquiring partial binary mode characteristic of video image
CN101551823A (en) * 2009-04-20 2009-10-07 浙江师范大学 Comprehensive multi-feature image retrieval method
CN102254165A (en) * 2011-08-12 2011-11-23 北方工业大学 Hand back vein identification method based on fusion of structural coding features and texture coding features
CN102662949A (en) * 2012-02-27 2012-09-12 安科智慧城市技术（中国）有限公司 Method and system for retrieving specified object based on multi-feature fusion
CN103440646A (en) * 2013-08-19 2013-12-11 成都品果科技有限公司 Similarity obtaining method for color distribution and texture distribution image retrieval

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR100750424B1 (en) * 2004-03-03 2007-08-21 닛본 덴끼 가부시끼가이샤 Image similarity calculation system, image search system, image similarity calculation method, and image similarity calculation program
CA2652714A1 (en) * 2006-05-29 2007-12-06 Philip Ogunbona Content based image retrieval
CN101789005A (en) * 2010-01-22 2010-07-28 深圳创维数字技术股份有限公司 Image searching method based on region of interest (ROI)
CN102663754B (en) * 2012-04-17 2014-12-10 北京博研新创数码科技有限公司 Image matching calculation method based on regional Gaussian weighting
CN103207910B (en) * 2013-04-08 2017-02-08 河南大学 Image retrieval method based on hierarchical features and genetic programming relevance feedback

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20080166026A1 (en) * 2007-01-10 2008-07-10 Samsung Electronics Co., Ltd. Method and apparatus for generating face descriptor using extended local binary patterns, and method and apparatus for face recognition using extended local binary patterns
CN101425136A (en) * 2008-10-24 2009-05-06 北京工业大学 Method and device for acquiring partial binary mode characteristic of video image
CN101551823A (en) * 2009-04-20 2009-10-07 浙江师范大学 Comprehensive multi-feature image retrieval method
CN102254165A (en) * 2011-08-12 2011-11-23 北方工业大学 Hand back vein identification method based on fusion of structural coding features and texture coding features
CN102662949A (en) * 2012-02-27 2012-09-12 安科智慧城市技术（中国）有限公司 Method and system for retrieving specified object based on multi-feature fusion
CN103440646A (en) * 2013-08-19 2013-12-11 成都品果科技有限公司 Similarity obtaining method for color distribution and texture distribution image retrieval

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111461147A (en) * 2020-04-30 2020-07-28 柳州智视科技有限公司 Binary coding organization algorithm based on image characteristics
CN111861981A (en) * 2020-05-29 2020-10-30 北京工商大学 Image color space fat point detection and evaluation method based on point, line and surface
CN111861981B (en) * 2020-05-29 2023-05-30 北京工商大学 Image color space fat point detection and evaluation method based on dotted line and plane
CN111985559A (en) * 2020-08-19 2020-11-24 合肥工业大学 Tire pattern structure similarity detection method based on boundary characteristics
CN114445659A (en) * 2022-01-20 2022-05-06 南京中医药大学 Method for identifying authenticity of spina date seeds based on image color and texture characteristics

Also Published As

Publication number Publication date
CN103440646A (en) 2013-12-11
CN103440646B (en) 2016-08-10

Similar Documents

Publication Publication Date Title
WO2015024383A1 (en) Similarity acquisition method for colour distribution and texture distribution image retrieval
CN108492343B (en) Image synthesis method for training data for expanding target recognition
CN108052980B (en) Image-based air quality grade detection method
WO2019163985A1 (en) Feature representation device, feature representation method, and program
WO2016127883A1 (en) Image area detection method and device
TWI338265B (en) System, apparatus, method and program for processing image
US9749503B2 (en) Image processing device, image processing method and recording medium
CN103065153A (en) Video key frame extraction method based on color quantization and clusters
CN103366178A (en) Method and device for carrying out color classification on target image
CN105069042A (en) Content-based data retrieval methods for unmanned aerial vehicle spying images
CN106250431B (en) A kind of Color Feature Extraction Method and costume retrieval system based on classification clothes
CN103530638A (en) Method for matching pedestrians under multiple cameras
CN104462481A (en) Comprehensive image retrieval method based on colors and shapes
CN105404657A (en) CEDD feature and PHOG feature based image retrieval method
CN108805139B (en) Image similarity calculation method based on frequency domain visual saliency analysis
US8995759B2 (en) Image processing apparatus and image processing method
CN109886281A (en) One kind is transfinited learning machine color image recognition method based on quaternary number
WO2017215194A1 (en) Image processing method and device and storage medium thereof
CN109213886A (en) Image search method and system based on image segmentation and Fuzzy Pattern Recognition
CN105374010A (en) A panoramic image generation method
Garg et al. Color based segmentation using K-mean clustering and watershed segmentation
CN108648245B (en) Information extraction method and device for well logging interpretation curve
JP6151908B2 (en) Learning device, identification device, and program thereof
JP6789175B2 (en) Image recognizers, methods, and programs
Sidqi et al. Image classification using K–mean algorithm

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: 14837926

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: 14837926

Country of ref document: EP

Kind code of ref document: A1

122 Ep: pct application non-entry in european phase

Ref document number: 14837926

Country of ref document: EP

Kind code of ref document: A1

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 16.11.2016)