CN105701150A

CN105701150A - Intuitionistic fuzzy similarity degree based image retrieving method and system

Info

Publication number: CN105701150A
Application number: CN201511016451.5A
Authority: CN
Inventors: 王子睿; 王志慧; 赵艺群; 许文伯; 樊月娇; 萨齐拉; 李锦林
Original assignee: Inner Mongolia University
Current assignee: Inner Mongolia University
Priority date: 2015-12-29
Filing date: 2015-12-29
Publication date: 2016-06-22

Abstract

The invention discloses an image retrieval method based on intuitionistic fuzzy similarity and its system. The method includes: step 1, performing intuitionistic fuzzification on images in an image library to obtain an intuitionistic fuzzy set matrix model of the image; step 2, according to The intuitionistic fuzzy set matrix model calculates the intuitionistic fuzzy similarity between each image in the image database and the retrieved image; and step 3, outputs the retrieved images in order from large to small according to the intuitionistic fuzzy similarity to obtain the final retrieval result . The invention can effectively retrieve images to be retrieved, is simple and fast, and has better retrieval effect for images with simple backgrounds.

Description

An image retrieval method and system based on intuitionistic fuzzy similarity

技术领域technical field

本发明涉及图像内容检索技术，特别是涉及一种基于直觉模糊相似度的图像检索方法及其系统。The invention relates to image content retrieval technology, in particular to an image retrieval method and system based on intuitionistic fuzzy similarity.

背景技术Background technique

当今世界科技高速发展，随着多媒体设备的普及，图像也得到了越来越广泛的应用，大量图像被存放到图像数据库中，图像数据库也越来越大，图像较之于普通的文本，包含有更加丰富的信息，如颜色、形状、纹理、位置、环境等。同时，图像的信息组织具有无序性，如何实现对图像进行有效的、统一的管理和查询一直是图像处理领域的一项研究热点。图像检索技术就是为实现在大量图片数据中快速有效地检索出所需要的信息。With the rapid development of science and technology in today's world, with the popularization of multimedia equipment, images have been more and more widely used. A large number of images are stored in the image database, and the image database is also growing. Compared with ordinary text, images contain There are richer information such as color, shape, texture, location, environment, etc. At the same time, the information organization of images is disorderly. How to realize effective and unified management and query of images has always been a research hotspot in the field of image processing. Image retrieval technology is to quickly and efficiently retrieve the required information from a large amount of image data.

图像检索通常分为两种方法：基于文本的图像检索(Text-basedImageRetrieval,TBIR)和基于内容的图像检索(Content-basedImageRetrieval,CBIR)。TBIR是基于比对关键字和图片标签的匹配度大小来检索的，该方法易于理解，实现简单，但它过于依赖人对图片的主观感知并且注解工作量大，该方法的基本流程为首先对图像库中的图像进行人工标注，得到图像文本数据，用户使用图像关键字对图像文本数据进行文本检索，得到结果输出。Image retrieval is usually divided into two methods: Text-based Image Retrieval (TBIR) and Content-based Image Retrieval (CBIR). TBIR is retrieved based on the matching degree of keywords and image tags. This method is easy to understand and simple to implement, but it relies too much on people's subjective perception of images and the annotation workload is heavy. The basic process of this method is to first The images in the image database are manually marked to obtain the image text data, and the user uses the image keywords to perform text retrieval on the image text data to obtain the output result.

CBIR不同于TBIR的精确匹配，采用相似匹配，并融合了计算机视觉、图像处理、图像理解和数据库等多个领域的技术成果，以避免人工描述的主观性，它允许用户输入一张图片，以查找相似内容的图片。CBIR主要利用图像包含的颜色、纹理、形状、空间位置以及它们的组合特征建立特征索引库，并通过图像特征向量特征之间的近似度量进行相似性索引。常用的CBIR通常分为基于全局特征和基于图像局部信息的检索方法，基于图像局部信息的检索方法对图像进行区域分割，选择感兴趣的区域进行特征提取。区域分割试图通过物体层次来表示图像的内容，理想的图像分割可以把图像分割成为语义区域，从而提取高水平的语义特征，但是，现有的分割技术并不能实现对图像中对象的精确分割，只能实现大概的分割，很难达到理想的分割状态。CBIR is different from the exact matching of TBIR. It uses similar matching and integrates technical achievements in multiple fields such as computer vision, image processing, image understanding and database to avoid the subjectivity of manual description. It allows users to input a picture to Find images with similar content. CBIR mainly uses the color, texture, shape, spatial position and their combination features contained in the image to establish a feature index library, and performs similarity indexing through the approximate measure between image feature vector features. The commonly used CBIR is usually divided into retrieval methods based on global features and image local information. The retrieval method based on image local information performs region segmentation on the image and selects the region of interest for feature extraction. Region segmentation attempts to represent the content of the image through the object level. The ideal image segmentation can segment the image into semantic regions, thereby extracting high-level semantic features. However, the existing segmentation technology cannot achieve accurate segmentation of objects in the image. Only approximate segmentation can be achieved, and it is difficult to achieve an ideal segmentation state.

图像底层如纹理、颜色、形状、语义等特征是CBIR所检索的主要内容，如何有效地提取和利用这些特征是CBIR的一项研究难点，基于模糊集理论，这些底层信息通常都携带一定的模糊性，因此，利用模糊集理论进行图像检索十分合理，直觉模糊集理论是人工智能科学的新领域，是模糊集的一种优化，目前，基于直觉模糊集理论的图像检索算法研究较少，且多数集中在使用直觉模糊集理论对图像间距离进行度量，算法较为冗杂。The underlying features of the image, such as texture, color, shape, and semantics, are the main content retrieved by CBIR. How to effectively extract and utilize these features is a research difficulty of CBIR. Based on fuzzy set theory, these underlying information usually carry a certain degree of fuzziness. Therefore, it is very reasonable to use fuzzy set theory for image retrieval. Intuitionistic fuzzy set theory is a new field of artificial intelligence science and an optimization of fuzzy sets. At present, there are few researches on image retrieval algorithms based on intuitionistic fuzzy set theory, and Most of them focus on using intuitionistic fuzzy set theory to measure the distance between images, and the algorithm is relatively complicated.

目前，基于直觉模糊集理论的图像检索算法面临如下问题：At present, image retrieval algorithms based on intuitionistic fuzzy set theory face the following problems:

(1)研究成果较少，直觉模糊集的优良性能在图像检索领域未得到充分利用。(1) There are few research results, and the excellent performance of intuitionistic fuzzy sets has not been fully utilized in the field of image retrieval.

(2)多数算法更倾向于使用图像间的距离度量，较为冗杂。(2) Most algorithms prefer to use the distance measure between images, which is more complicated.

发明内容Contents of the invention

本发明的目的在于提供一种基于直觉模糊相似度的图像检索方法及其系统，用于实现有效、简便快捷地检索出待检索图像。The purpose of the present invention is to provide an image retrieval method based on intuitionistic fuzzy similarity and its system, which are used to retrieve images to be retrieved effectively, conveniently and quickly.

为了实现上述目的，本发明提供一种基于直觉模糊相似度的图像检索方法，包括：In order to achieve the above object, the present invention provides a kind of image retrieval method based on intuitionistic fuzzy similarity, comprising:

步骤一，对图像库中的图像进行直觉模糊化，得到图像的直觉模糊集矩阵模型；Step 1, performing intuitionistic fuzzification on the images in the image library to obtain the intuitionistic fuzzy set matrix model of the image;

步骤二，根据直觉模糊集矩阵模型求取图像库中每一张图像与被检索图像的直觉模糊相似度；Step 2, calculating the intuitionistic fuzzy similarity between each image in the image database and the retrieved image according to the intuitionistic fuzzy set matrix model;

步骤三，将检索图像按照直觉模糊相似度从大到小依序输出，得到最终检索结果。Step 3: Output the retrieved images in descending order according to the intuitionistic fuzzy similarity to obtain the final retrieval result.

所述的图像检索方法，其中，所述步骤一中，包括：在进行直觉模糊化之前，对图像库中的图像进行预处理的步骤。The image retrieval method, wherein, in the first step, it includes the step of preprocessing the images in the image database before performing intuitive blurring.

所述的图像检索方法，其中，所述步骤一中，包括：The image retrieval method, wherein, in the first step, including:

步骤1.1：将图像库中每张灰度图像矩阵中的像素值做归一化，并以如下方式计算图像的隶属度矩阵：Step 1.1: Normalize the pixel values in each grayscale image matrix in the image library, and calculate the membership matrix of the image in the following way:

$μ μ = = {Σ Σ}_{i i = = 11}^{M m} {Σ Σ}_{j j = = 11}^{N N} {x x}_{i i j j} / / ((M m \times \times N N)),, β β = = 11 / / (({f f}_{m m a a x x} - - {f f}_{m m i i n no})) - - - - - - ((11))$

${μ μ}_{i i j j} (({x x}_{i i j j})) = = (({f f}_{m m a a x x} - - {f f}_{m m i i n no})) \cdot &Center Dot; exp exp ((- - (({f f}_{m m a a x x} - - {f f}_{m m i i n no})) \cdot \cdot | | {x x}_{i i j j} - - μ μ | |)) = = \frac{11}{β β} \cdot &Center Dot; exp exp ((- - \frac{{x x}_{i i j j} - - μ μ}{β β})) - - - - - - ((22))$

其中，i＝1,2,…M，j＝1,2,…N，μ_ij(x_ij)为像素点x_ij的隶属度，f_max和f_min分别为归一化后的图像矩阵中的最大值和最小值，M、N分别为图像矩阵的行值和列值；Among them, i=1,2,...M, j=1,2,...N, μ _ij (x _ij ) is the membership degree of pixel point x _ij , f _max and f _min are the normalized image matrix The maximum value and the minimum value of , M, N are the row value and column value of image matrix respectively;

步骤1.2：在得到图像的隶属度矩阵后，以如下方式计算图像的非隶属度矩阵：Step 1.2: After obtaining the membership degree matrix of the image, calculate the non-membership degree matrix of the image as follows:

${&upsi; &upsi;}_{i i j j} (({x x}_{i i j j})) = = \frac{11 - - {μ μ}_{i i j j} (({x x}_{i i j j}))}{11 + + {λμ λμ}_{i i j j} (({x x}_{i i j j}))} - - - - - - ((33))$

其中，υ_ij(x_ij)为像素点x_ij的非隶属度，λ＞0。Among them, υ _ij ( _{xij ) is the non-membership degree of pixel point x ij} _, λ>0.

所述的图像检索方法，其中，所述步骤二中，包括：The image retrieval method, wherein, in the second step, includes:

以如下方式求图像库中每张图像与被检索图像的直觉模糊相似度：Find the intuitionistic fuzzy similarity between each image in the image library and the retrieved image in the following way:

$\begin{matrix} θ θ (({A A}_{11},, {A A}_{22})) \\ = = 11 - - {Σ Σ}_{j j = = 11}^{n no} {ω ω}_{j j} [[\frac{| | {μ μ}_{{A A}_{11}} (({x x}_{j j})) - - {μ μ}_{{A A}_{22}} (({x x}_{j j})) | | + + | | {&upsi; &upsi;}_{{A A}_{11}} (({x x}_{j j})) - - {&upsi; &upsi;}_{{A A}_{22}} (({x x}_{j j})) | |}{44} \\ + + \frac{max max {{| | {μ μ}_{{A A}_{11}} (({x x}_{j j})) - - {μ μ}_{{A A}_{22}} (({x x}_{j j})) | |,, | | {&upsi; &upsi;}_{{A A}_{11}} (({x x}_{j j})) - - {&upsi; &upsi;}_{{A A}_{22}} (({x x}_{j j})) | |}}}{22}]] \end{matrix} - - - - - - ((44))$

其中，ω_j＝(1/n，1/n，·..，1/n)^T为权值，n为被检索图像的像素点总数，A₁和A₂分别表示被检索图像的直觉模糊集矩阵模型和检索图像库中任意一张图像的直觉模糊集矩阵模型，A_i(j)为直觉模糊集矩阵模型A_i的第j个坐标点的值，θ(A₁,A₂)表示A₁和A₂分别对应的原图像的直觉模糊相似度。Among them, ω _j = (1/n, 1/n, .., 1/n) ^T is the weight, n is the total number of pixels of the retrieved image, A ₁ and A ₂ represent the intuitive blur of the retrieved image Set matrix model and retrieve the intuitionistic fuzzy set matrix model of any image in the image database, A _i (j) is the value of the jth coordinate point of the intuitionistic fuzzy set matrix model A _i , θ(A ₁ , A ₂ ) represents the intuitionistic fuzzy similarity of the original image corresponding to A ₁ and A ₂ respectively.

所述的图像检索方法，其中，所述步骤三中，包括：The image retrieval method, wherein, in the third step, includes:

选取与检索图像直觉模糊相似度最大的图像作为所求的被检索图像。Select the image with the largest intuitionistic fuzzy similarity with the retrieved image as the retrieved image.

为了实现上述目的，本发明提供一种基于直觉模糊相似度的图像检索系统，包括：In order to achieve the above object, the present invention provides an image retrieval system based on intuitionistic fuzzy similarity, comprising:

模糊化处理模块，用于对图像库中的图像进行直觉模糊化，得到图像的直觉模糊集矩阵模型；The fuzzy processing module is used to carry out intuitionistic fuzzy to the image in the image library, and obtains the intuitionistic fuzzy set matrix model of the image;

相似度获取模块，用于根据直觉模糊集矩阵模型求取图像库中每一张图像与被检索图像的直觉模糊相似度；A similarity acquisition module is used to obtain the intuitionistic fuzzy similarity between each image in the image library and the retrieved image according to the intuitionistic fuzzy set matrix model;

检索结果获取模块，用于将检索图像按照直觉模糊相似度从大到小依序输出，得到最终检索结果。The retrieval result acquisition module is used to sequentially output the retrieved images according to the intuitionistic fuzzy similarity from large to small to obtain the final retrieval result.

所述的图像检索系统，其中，还包括：预处理模块，用于在进行直觉模糊化之前，对图像库中的图像进行预处理。The image retrieval system further includes: a preprocessing module, used to preprocess the images in the image library before performing intuitional blurring.

所述的图像检索系统，其中，所述模糊化处理模块进一步包括：The image retrieval system, wherein, the blur processing module further includes:

隶属度矩阵模块，用于将图像库中每张灰度图像矩阵中的像素值做归一化，并以如下方式计算图像的隶属度矩阵；The membership matrix module is used to normalize the pixel values in each grayscale image matrix in the image library, and calculate the membership matrix of the image in the following manner;

${μ μ}_{i i j j} (({x x}_{i i j j})) = = (({f f}_{m m a a x x} - - {f f}_{m m i i n no})) \cdot &Center Dot; exp exp ((- - (({f f}_{m m a a x x} - - {f f}_{m m i i n no})) \cdot &Center Dot; | | {x x}_{i i j j} - - μ μ | |)) = = \frac{11}{β β} \cdot &Center Dot; exp exp ((- - \frac{{x x}_{i i j j} - - μ μ}{β β})) - - - - - - ((22))$

非隶属度矩阵模块，用于在得到图像的隶属度矩阵后，以如下方式计算图像的非隶属度矩阵：The non-membership matrix module is used to calculate the non-membership matrix of the image in the following manner after obtaining the membership matrix of the image:

所述的图像检索系统，其中，所述相似度获取模块以如下方式求图像库中每张图像与被检索图像的直觉模糊相似度：The image retrieval system, wherein, the similarity acquisition module seeks the intuitionistic fuzzy similarity between each image in the image library and the retrieved image in the following manner:

其中，ω_j＝(1/n，1/n，，1/n)^T为权值，n为被检索图像的像素点总数，A₁和A₂分别表示被检索图像的直觉模糊集矩阵模型和检索图像库中任意一张图像的直觉模糊集矩阵模型，A_i(j)为直觉模糊集矩阵模型A_i的第j个坐标点的值，θ(A₁,A₂)表示A₁和A₂分别对应的原图像的直觉模糊相似度。Among them, ω _j = (1/n, 1/n,, 1/n) ^T is the weight, n is the total number of pixels of the retrieved image, A ₁ and A ₂ respectively represent the intuitionistic fuzzy set matrix model of the retrieved image and retrieve the intuitionistic fuzzy set matrix model of any image in the image library, A _i (j) is the value of the jth coordinate point of the intuitionistic fuzzy set matrix model A _i , θ(A ₁ , A ₂ ) represents the intuitionistic fuzzy similarity of the original image corresponding to A ₁ and A ₂ respectively.

所述的图像检索系统，其中，所述检索结果获取模块选取与检索图像直觉模糊相似度最大的图像作为所求的被检索图像。In the image retrieval system, the retrieval result acquisition module selects the image with the largest intuitional fuzzy similarity with the retrieved image as the retrieved image.

与现有技术相比，本发明的有益技术效果是：Compared with the prior art, the beneficial technical effect of the present invention is:

本发明为了解决现有技术所存在的上述技术问题，提供了一种可有效提高检索精度及效率的基于直觉模糊理论的图像检索方法，该方法通过Gamma函数和Sugeno模糊补集来构造图像的直觉模糊集矩阵模型，测量图像的直觉模糊相似度，并用其进行图像检索。实验表明：本发明提出的算法可以有效地检索出待检索图像，且简便快捷，对于背景简单的图像检索效果更好。In order to solve the above-mentioned technical problems existing in the prior art, the present invention provides an image retrieval method based on intuitionistic fuzzy theory that can effectively improve retrieval accuracy and efficiency. The method uses Gamma function and Sugeno fuzzy complement to construct an image intuition The fuzzy set matrix model measures the intuitionistic fuzzy similarity of images and uses it for image retrieval. Experiments show that the algorithm proposed by the invention can effectively retrieve images to be retrieved, and is simple and fast, and has a better retrieval effect on images with simple backgrounds.

附图说明Description of drawings

图1为本发明基于直觉模糊相似度的图像检索方法流程图。FIG. 1 is a flowchart of an image retrieval method based on intuitionistic fuzzy similarity in the present invention.

图2为本发明的花实例检索效果图。Fig. 2 is a flower instance retrieval effect diagram of the present invention.

图3为本发明的公交车实例检索效果图。Fig. 3 is a retrieval effect diagram of the bus example of the present invention.

图4为本发明的大象实例检索效果图。Fig. 4 is an effect diagram of elephant instance retrieval in the present invention.

图5为本发明基于直觉模糊相似度的图像检索系统结构图。FIG. 5 is a structural diagram of an image retrieval system based on intuitionistic fuzzy similarity in the present invention.

具体实施方式detailed description

以下结合附图和具体实施例对本发明进行详细描述，但不作为对本发明的限定。The present invention will be described in detail below in conjunction with the accompanying drawings and specific embodiments, but not as a limitation of the present invention.

如图1所示，为本发明基于直觉模糊相似度的图像检索方法流程图。该流程图描述了一种基于直觉模糊理论的图像检索方法，按照如下步骤进行：As shown in FIG. 1 , it is a flow chart of the image retrieval method based on intuitionistic fuzzy similarity of the present invention. This flow chart describes an image retrieval method based on intuitionistic fuzzy theory, which proceeds as follows:

步骤1：对图像库中的图像进行预处理。Step 1: Preprocess the images in the image library.

步骤2：对预处理后的图像进行直觉模糊化，得到每张图像的直觉模糊集矩阵模型。Step 2: Perform intuitionistic fuzzification on the preprocessed image to obtain the intuitionistic fuzzy set matrix model of each image.

步骤3：求图像库中每一张图像与被检索图像的直觉模糊相似度。Step 3: Calculate the intuitionistic fuzzy similarity between each image in the image database and the retrieved image.

步骤4：将所得的检索图像按照直觉模糊相似度从大到小依序输出，得到最终检索结果。Step 4: Output the retrieved images in order according to the intuitionistic fuzzy similarity from large to small to obtain the final retrieval result.

所述步骤1如下：Said step 1 is as follows:

步骤1.1：对图像库中的图像使用均值滤波器滤波。Step 1.1: Filter the images in the image library using the mean filter.

步骤1.2：对滤波后的图像进行灰度化处理，得到灰度图像矩阵。Step 1.2: Perform grayscale processing on the filtered image to obtain a grayscale image matrix.

所述步骤2如下：Said step 2 is as follows:

步骤2.1：求图像的隶属度矩阵。Step 2.1: Find the membership degree matrix of the image.

将每张灰度图像矩阵中的像素值做归一化，然后使用Gamma函数来求取图像的模糊隶属度，Gamma函数的定义如下：Normalize the pixel values in each grayscale image matrix, and then use the Gamma function to obtain the fuzzy membership of the image. The Gamma function is defined as follows:

$G G ((x x)) = = \frac{{((x x - - m m))}^{γ γ - - 11} exp exp ((- - \frac{x x - - m m}{β β}))}{{β β}^{γ γ} Γ Γ ((γ γ))} - - - - - - ((11))$

其中，γ为形状参数，m为位置参数，β为尺度参数，Γ(1)＝1。Among them, γ is a shape parameter, m is a position parameter, β is a scale parameter, and Γ(1)=1.

当m≠0且γ＝1时，Gamma函数可以简化为：When m≠0 and γ=1, the Gamma function can be simplified as:

$G G ((x x)) = = \frac{11}{β β} exp exp ((- - \frac{x x - - m m}{β β})) - - - - - - ((22))$

将图像做归一化，然后，在此基础上，求出图像的隶属度矩阵：Normalize the image, and then, on this basis, find the membership matrix of the image:

$μ μ = = {Σ Σ}_{i i = = 11}^{M m} {Σ Σ}_{j j = = 11}^{N N} {x x}_{i i j j} / / ((M m \times \times N N)),, β β = = 11 / / (({f f}_{m m a a x x} - - {f f}_{m m i i n no})) - - - - - - ((33))$

其中，i＝1,2,…M，j＝1,2,…N，f_max和f_min分别为归一化后的图像矩阵f(x)中的最大值和最小值。M、N分别为图像矩阵的行值和列值，M、N随图像矩阵的行数和列数变化而变化，无特定的取值范围。Wherein, i=1,2,...M, j=1,2,...N, f _max and f _min are the maximum value and minimum value in the normalized image matrix f(x) respectively. M and N are row values and column values of the image matrix respectively, and M and N vary with the number of rows and columns of the image matrix, and there is no specific value range.

进而，可以得到像素点x_ij的隶属度矩阵为：Furthermore, the membership degree matrix of pixel x _ij can be obtained as:

${μ μ}_{i i j j} (({x x}_{i i j j})) = = (({f f}_{m m a a x x} - - {f f}_{m m i i n no})) \cdot &Center Dot; exp exp ((- - (({f f}_{m m a a x x} - - {f f}_{m m i i n no})) \cdot &Center Dot; | | {x x}_{i i j j} - - μ μ | |)) = = \frac{11}{β β} \cdot \cdot exp exp ((- - \frac{{x x}_{i i j j} - - μ μ}{β β})) - - - - - - ((44))$

其中，μ_ij(x_ij)为像素点x_ij的隶属度，是基于Gamma函数代换、化简、变换而得到的。Among them, μ _ij ( _xij ) is the membership degree of pixel point x _ij , which is obtained based on Gamma function substitution, simplification, and transformation.

步骤2.2：求图像的非隶属度矩阵。Step 2.2: Find the non-membership degree matrix of the image.

得到图像的隶属度矩阵后，使用Sugeno模糊补集来求取图像的非隶属度矩阵，Sugeno补集定义如下：After obtaining the membership matrix of the image, use the Sugeno fuzzy complement to obtain the non-membership matrix of the image. The Sugeno complement is defined as follows:

$N N ((g g ((x x)))) = = \frac{11 - - g g ((x x))}{11 + + λ λ g g ((x x))} - - - - - - ((55))$

其中，N(1)＝0且N(0)＝1。Wherein, N(1)=0 and N(0)=1.

计算图像的非隶属度矩阵如下：Calculate the non-membership matrix of the image as follows:

${&upsi; &upsi;}_{i i j j} (({x x}_{i i j j})) = = \frac{11 - - {μ μ}_{i i j j} (({x x}_{i i j j}))}{11 + + {λμ λμ}_{i i j j} (({x x}_{i i j j}))} - - - - - - ((66))$

其中，υ_ij(x_ij)为像素点x_ij的非隶属度，μ_ij(x_ij)为像素点x_ij的隶属度，λ＞0，本发明取λ＝0.5。Wherein, υ _ij ( _xij ) is the non-membership degree of pixel point x _ij , μ _ij ( _xij ) is the membership degree of pixel point x _ij , λ>0, and λ=0.5 in the present invention.

上述得到的隶属度矩阵、非隶属度矩阵为直觉模糊集矩阵模型。所述步骤3如下：The membership degree matrix and non-membership degree matrix obtained above are intuitionistic fuzzy set matrix models. Said step 3 is as follows:

以如下方式求图像库中每一张图像与被检索图像的直觉模糊相似度：Calculate the intuitionistic fuzzy similarity between each image in the image database and the retrieved image in the following way:

$\begin{matrix} θ θ (({A A}_{11},, {A A}_{22})) \\ = = 11 - - {Σ Σ}_{j j = = 11}^{n no} {ω ω}_{j j} [[\frac{| | {μ μ}_{{A A}_{11}} (({x x}_{j j})) - - {μ μ}_{{A A}_{22}} (({x x}_{j j})) | | + + | | {&upsi; &upsi;}_{{A A}_{11}} (({x x}_{j j})) - - {&upsi; &upsi;}_{{A A}_{22}} (({x x}_{j j})) | |}{44} \\ + + \frac{max max {{| | {μ μ}_{{A A}_{11}} (({x x}_{j j})) - - {μ μ}_{{A A}_{22}} (({x x}_{j j})) | |,, | | {&upsi; &upsi;}_{{A A}_{11}} (({x x}_{j j})) - - {&upsi; &upsi;}_{{A A}_{22}} (({x x}_{j j})) | |}}}{22}]] \end{matrix} - - - - - - ((77))$

所述步骤4如下：Said step 4 is as follows:

将所得的检索图像按照直觉模糊相似度从大到小依序输出，得到最终检索结果，选取与检索图像直觉模糊相似度最大的图像即为所求的被检索的图像。The obtained retrieved images are output in descending order according to the intuitionistic fuzzy similarity to obtain the final retrieval result, and the image with the largest intuitionistic fuzzy similarity to the retrieved image is selected as the retrieved image.

下面将通过一具体实施方式来进一步说明本发明技术方案的技术效果。The technical effect of the technical solution of the present invention will be further described below through a specific implementation manner.

为了验证本发明算法的性能，使用具有1000张通用图像的图片库对提出的基于直觉模糊理论的检索算法进行实验。下面分别给出了三个图像检索实例，检索图像分别为：花、公交车、大象。检索结果如图2、图3、图4所示。In order to verify the performance of the algorithm of the present invention, the proposed retrieval algorithm based on intuitionistic fuzzy theory is tested using a picture library with 1000 general images. Three examples of image retrieval are given below, and the retrieved images are: flowers, buses, and elephants. The retrieval results are shown in Figure 2, Figure 3, and Figure 4.

图像检索的通用效果评价是查全率和查准率，查全率和查准率的定义如下：The general effect evaluation of image retrieval is recall rate and precision rate. The definitions of recall rate and precision rate are as follows:

查全率＝检索到的相关图像数目/所有相关图像数目Recall rate = number of relevant images retrieved/number of all relevant images

查准率＝检索到的相关图像数目/已检索出的图像数目Precision = number of relevant images retrieved/number of images retrieved

本发明从1000张图像中选取10类，每类随机选取10张图像，构成一个具有100张图像的图像集，对每类图像的所有图像都进行检索实验，并求其检索的查全率和查准率，最后取平均值，作为该类图像的查全率和查准率，得到的结果如表1所示。The present invention selects 10 categories from 1000 images, each category randomly selects 10 images to form an image collection with 100 images, and performs a retrieval experiment on all images of each category of images, and seeks the retrieval recall rate and The precision rate, and finally take the average value as the recall rate and precision rate of this type of image, and the results are shown in Table 1.

表1算法的结果评价Table 1 Evaluation of the results of the algorithm

类别category 图像image 查全率recall 查准率Precision 11 人物figure 0.140.14 0.140.14 22 沙滩beach 0.220.22 0.220.22 33 建筑architecture 0.20.2 0.20.2 44 公交车bus 0.210.21 0.210.21 55 恐龙Dinosaur 0.790.79 0.790.79 66 大象elephant 0.160.16 0.160.16 77 花flower 0.350.35 0.350.35 88 马horse 0.340.34 0.340.34 99 雪山snow mountain 0.210.21 0.210.21 1010 食物food 0.160.16 0.160.16

从实施例的结果可以显著地表示出，本发明提出的算法得到了可观的检索结果，除了可以检索出待检索的图像外，也可以有效地检索出部分同类的图像，并且可以看出，对于例如恐龙这种背景简单的图像，查全率和查准率都比较高，说明对于背景较为简单的这类图像，本发明的检索算法非常有效。It can be clearly shown from the results of the embodiments that the algorithm proposed by the present invention has obtained considerable retrieval results. In addition to retrieving the image to be retrieved, it can also effectively retrieve some images of the same type, and it can be seen that for For example, images with simple backgrounds such as dinosaurs have relatively high recall and precision, which shows that the retrieval algorithm of the present invention is very effective for such images with relatively simple backgrounds.

如图5所示，为本发明的基于直觉模糊相似度的图像检索系统结构图。该系统500与图1所述的方法相对应，结合图1，该系统500包括如下模块：As shown in FIG. 5 , it is a structural diagram of the image retrieval system based on intuitionistic fuzzy similarity of the present invention. The system 500 corresponds to the method described in FIG. 1. In conjunction with FIG. 1, the system 500 includes the following modules:

预处理模块51，用于对图像库中的图像进行预处理。The preprocessing module 51 is used for preprocessing the images in the image database.

模糊化处理模块52，用于对预处理后的图像进行直觉模糊化，得到每张图像的直觉模糊集矩阵模型。The fuzzy processing module 52 is configured to perform intuitionistic fuzzy on the preprocessed image to obtain an intuitionistic fuzzy set matrix model of each image.

相似度获取模块53，用于获取图像库中每一张图像与被检索图像的直觉模糊相似度。以及The similarity acquisition module 53 is used to acquire the intuitionistic fuzzy similarity between each image in the image database and the retrieved image. as well as

检索结果获取模块54，用于将所得的检索图像按照相似度从大到小依序输出，得到最终检索结果。The retrieval result acquisition module 54 is configured to output the obtained retrieval images in descending order of similarity to obtain the final retrieval result.

所述预处理模块51进一步包括：The preprocessing module 51 further includes:

滤波模块511，用于对图像库中的图像使用均值滤波器滤波。The filtering module 511 is configured to use a mean value filter to filter the images in the image library.

灰度化模块512，用于对滤波后的图像进行灰度化处理，得到灰度图像矩阵。The grayscale module 512 is configured to perform grayscale processing on the filtered image to obtain a grayscale image matrix.

所述模糊化处理模块52进一步包括：The fuzzy processing module 52 further includes:

隶属度矩阵模块521，用于求图像的隶属度矩阵。The membership degree matrix module 521 is used to calculate the membership degree matrix of the image.

进而，可以得到像素点x_ij的隶属度为：Furthermore, the degree of membership of the pixel point x _ij can be obtained as:

${μ μ}_{i i j j} (({x x}_{i i j j})) = = (({f f}_{m m a a x x} - - {f f}_{m m i i n no})) \cdot &Center Dot; exp exp ((- - (({f f}_{m m a a x x} - - {f f}_{m m i i n no})) \cdot &Center Dot; | | {x x}_{i i j j} - - μ μ | |)) = = \frac{11}{β β} \cdot &Center Dot; exp exp ((- - \frac{{x x}_{i i j j} - - μ μ}{β β})) - - - - - - ((44))$

非隶属度矩阵模块522，用于求图像的非隶属度矩阵。The non-membership degree matrix module 522 is used to calculate the non-membership degree matrix of the image.

其中，N(1)＝0且N(0)＝1。Wherein, N(1)=0 and N(0)=1.

上述得到的隶属度矩阵、非隶属度矩阵为直觉模糊集矩阵模型。The membership degree matrix and non-membership degree matrix obtained above are intuitionistic fuzzy set matrix models.

所述相似度获取模块53采用以下方式获取图像库中每一张图像与被检索图像的直觉模糊相似度：The similarity acquisition module 53 acquires the intuitionistic fuzzy similarity between each image in the image library and the retrieved image in the following manner:

其中，ω_j＝(1/n，l/n，，1/n)^T为权值，n为被检索图像的像素点总数，A₁和A₂分别表示被检索图像的直觉模糊集矩阵模型和检索图像库中任意一张图像的直觉模糊集矩阵模型，A_i(j)为直觉模糊集矩阵模型A_i在第j个坐标点的值，θ(A₁,A₂)表示A₁和A₂分别对应的原图像的直觉模糊相似度。Among them, ω _j = (1/n, l/n,, 1/n) ^T is the weight, n is the total number of pixels of the retrieved image, A ₁ and A ₂ respectively represent the intuitionistic fuzzy set matrix model of the retrieved image and retrieve the intuitionistic fuzzy set matrix model of any image in the image library, A _i (j) is the value of the intuitionistic fuzzy set matrix model A _i at the jth coordinate point, θ(A ₁ , A ₂ ) represents the intuitionistic fuzzy similarity of the original image corresponding to A ₁ and A ₂ respectively.

所述检索结果获取模块54采用以下方式得到最终检索结果：The retrieval result acquisition module 54 obtains the final retrieval result in the following manner:

本发明提出了一种基于直觉模糊相似度的图像检索方法，其通过Gamma函数和Sugeno模糊补集来构造图像的直觉模糊集矩阵模型，测量图像的直觉模糊相似度，并用其进行图像检索。实验表明本发明提出的算法可以有效地检索出待检索图像，且简便快捷，对于背景简单的图像检索效果更好。The invention proposes an image retrieval method based on intuitionistic fuzzy similarity, which uses Gamma function and Sugeno fuzzy complement to construct an intuitionistic fuzzy set matrix model of an image, measures the intuitionistic fuzzy similarity of an image, and uses it to perform image retrieval. Experiments show that the algorithm proposed by the invention can effectively retrieve the image to be retrieved, and is simple and fast, and has a better retrieval effect on images with simple backgrounds.

当然，本发明还可有其它多种实施例，在不背离本发明精神及其实质的情况下，熟悉本领域的技术人员当可根据本发明做出各种相应的改变和变形，但这些相应的改变和变形都应属于本发明所附的权利要求的保护范围。Of course, the present invention can also have other various embodiments, and those skilled in the art can make various corresponding changes and deformations according to the present invention without departing from the spirit and essence of the present invention. All changes and deformations should belong to the protection scope of the appended claims of the present invention.

Claims

1. An image retrieval method based on intuitionistic fuzzy similarity is characterized by comprising the following steps:

firstly, carrying out intuitive fuzzification on images in an image library to obtain an intuitive fuzzy set matrix model of the images;

step two, calculating the intuitive fuzzy similarity between each image in the image library and the image to be searched according to the intuitive fuzzy set matrix model;

and step three, outputting the retrieval images in sequence from large to small according to the intuitive fuzzy similarity to obtain a final retrieval result.

2. The image retrieval method according to claim 1, wherein the first step comprises: and preprocessing the images in the image library before the intuitive fuzzification is carried out.

3. The image retrieval method according to claim 1 or 2, wherein the first step comprises:

step 1.1: normalizing the pixel value in each gray level image matrix in the image library, and calculating the membership degree matrix of the image in the following way:

μ = Σ_{i = 1}^{M} Σ_{j = 1}^{N} x_{i j} / (M \times N), β = 1 / (f_{m a x} - f_{m i n}) - - - (1)

μ_{i j} (x_{i j}) = (f_{m a x} - f_{m i n}) \cdot \exp (- (f_{m a x} - f_{m i n}) \cdot | x_{i j} - μ |) = \frac{1}{β} \cdot \exp (- \frac{x_{i j} - μ}{β}) - - - (2)

wherein, i is 1,2, … M, j is 1,2,…N，μ_ij(x_ij) Is a pixel point x_ijDegree of membership of f_maxAnd f_minRespectively, the maximum value and the minimum value in the normalized image matrix, and M, N respectively, the row value and the column value of the image matrix;

step 1.2: after obtaining the membership matrix of the image, calculating the non-membership matrix of the image in the following way:

{&upsi;}_{i j} (x_{i j}) = \frac{1 - μ_{i j} (x_{i j})}{1 + {λμ}_{i j} (x_{i j})} - - - (3)

wherein upsilon is_ij(x_ij) Is a pixel point x_ijλ > 0.

4. The image retrieval method according to claim 3, wherein the second step includes:

and solving the intuitionistic fuzzy similarity between each image in the image library and the searched image in the following way:

\begin{matrix} θ (A_{1}, A_{2}) \\ = 1 - Σ_{j = 1}^{n} ω_{j} [\frac{| μ_{A_{1}} (x_{j}) - μ_{A_{2}} (x_{j}) | + | {&upsi;}_{A_{1}} (x_{j}) - {&upsi;}_{A_{2}} (x_{j}) |}{4} \\ + \frac{\max {| μ_{A_{1}} (x_{j}) - μ_{A_{2}} (x_{j}) |, | {&upsi;}_{A_{1}} (x_{j}) - {&upsi;}_{A_{2}} (x_{j}) |}}{2}] \end{matrix} - - - (4)

wherein, ω is_j＝(1/n，1/n，···，1/n)^TIs a weight value, n is the total number of pixel points of the image to be searched, A₁And A₂Respectively representing an intuitive fuzzy set matrix model of the image to be searched and an intuitive fuzzy set matrix model of any one image in the search image library,A_i(j) set matrix model A is blurred for intuition_iOf the jth coordinate point of (a), theta (a)₁,A₂) Is represented by A₁And A₂The intuitive fuzzy similarity of the corresponding original images.

5. The image retrieval method according to claim 1,2 or 4, wherein in the third step, the method comprises:

and selecting the image with the maximum intuitive fuzzy similarity with the retrieval image as the searched image.

6. An image retrieval system based on intuitive fuzzy similarity, comprising:

the blurring processing module is used for performing intuitive blurring on the images in the image library to obtain an intuitive fuzzy set matrix model of the images;

the similarity acquisition module is used for calculating the intuitive fuzzy similarity between each image in the image library and the image to be searched according to the intuitive fuzzy set matrix model;

and the retrieval result acquisition module is used for sequentially outputting the retrieval images from large to small according to the intuitive fuzzy similarity to obtain a final retrieval result.

7. The image retrieval system of claim 6, further comprising: and the preprocessing module is used for preprocessing the images in the image library before the intuitive fuzzification is carried out.

8. The image retrieval system of claim 6 or 7, wherein the blurring processing module further comprises:

the membership matrix module is used for normalizing the pixel values in each gray level image matrix in the image library and calculating the membership matrix of the image in the following way;

μ = Σ_{i = 1}^{M} Σ_{j = 1}^{N} x_{i j} / (M \times N), β = 1 / (f_{m a x} - f_{m i n}) - - - (1)

μ_{i j} (x_{i j}) = (f_{m a x} - f_{m i n}) \cdot \exp (- (f_{m a x} - f_{m i n}) \cdot | x_{i j} - μ |) = \frac{1}{β} \cdot \exp (- \frac{x_{i j} - μ}{β}) - - - (2)

wherein, i is 1,2, … M, j is 1,2, … N, mu_ij(x_ij) Is a pixel point x_ijDegree of membership of f_maxAnd f_minRespectively, the maximum value and the minimum value in the normalized image matrix, and M, N respectively, the row value and the column value of the image matrix;

the non-membership matrix module is used for calculating the non-membership matrix of the image in the following mode after obtaining the membership matrix of the image:

{&upsi;}_{i j} (x_{i j}) = \frac{1 - μ_{i j} (x_{i j})}{1 + {λμ}_{i j} (x_{i j})} - - - (3)

wherein upsilon is_ij(x_ij) Is a pixel point x_ijλ > 0.

9. The image retrieval system of claim 8, wherein the similarity acquisition module finds the intuitive blurred similarity of each image in the image library to the retrieved image as follows:

\begin{matrix} θ (A_{1}, A_{2}) \\ = 1 - Σ_{j = 1}^{n} ω_{j} [\frac{| μ_{A_{1}} (x_{j}) - μ_{A_{2}} (x_{j}) | + | {&upsi;}_{A_{1}} (x_{j}) - {&upsi;}_{A_{2}} (x_{j}) |}{4} \\ + \frac{\max {| μ_{A_{1}} (x_{j}) - μ_{A_{2}} (x_{j}) |, | {&upsi;}_{A_{1}} (x_{j}) - {&upsi;}_{A_{2}} (x_{j}) |}}{2}] \end{matrix} - - - (4)

10. The image retrieval system according to claim 6, 7 or 9, wherein the retrieval result acquisition module selects an image having a maximum intuitive blur similarity with the retrieved image as the retrieved image.