CN112667842B - Image texture retrieval method - Google Patents
Image texture retrieval method Download PDFInfo
- Publication number
- CN112667842B CN112667842B CN202011603191.2A CN202011603191A CN112667842B CN 112667842 B CN112667842 B CN 112667842B CN 202011603191 A CN202011603191 A CN 202011603191A CN 112667842 B CN112667842 B CN 112667842B
- Authority
- CN
- China
- Prior art keywords
- image
- log
- feature
- optimal
- gabor
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Image Analysis (AREA)
Abstract
The application discloses an image texture retrieval method, which relates to the technical field of image processing in computer vision, and comprises the following steps: converting a target image to be retrieved from a spatial domain into a frequency domain; performing convolution operation on the transformed target image and a Log-Gabor filter; performing inverse Fourier transform on the convolution operation result to obtain a Log-Gabor response image; acquiring a maximum edge position octal mode LMEPOP characteristic vector of the Log-Gabor response image through a local texture characteristic LBP; and retrieving similar images from an image library according to the extracted feature vectors, wherein the feature vectors of all the images are stored in the image library. The method solves the problems of low retrieval accuracy and long time consumption caused by the existence of the direct current component in the prior art, and achieves the effects of improving the retrieval accuracy and shortening the time consumption of retrieval.
Description
Technical Field
The invention relates to an image texture retrieval method, and belongs to the technical field of image processing in computer vision.
Background
Images play a crucial role in almost every aspect of human life, such as pattern recognition, artificial intelligence, crime prevention, surveillance, news, medicine, etc., and when a large number of image sets are maintained, images need to be searched and retrieved according to user requests. Meanwhile, many techniques have been proposed for browsing and searching images in a huge image database. Among them, the content-based image retrieval system is very popular because it has relatively low requirements for people. However, the content-based image retrieval is performed by similar types of images and according to user needs, searches are performed on an image database, which analyzes the contents of pictures, such as color, texture, shape, etc., rather than any information about the images, such as text annotation, creation time or place, etc., i.e., texture features of the base image. Texture features can be described by statistical properties of the gray levels of the image pixels. Researchers continually optimize search algorithms in order to improve the accuracy of searches and shorten the search time. The working principle of the methods is divided into two steps: indexing and searching. In the indexing step, the visual content (features) of the images in the database. Extracted in the form of a feature vector and stored in a feature database. In the searching step, the feature vectors of the images are retrieved and compared with all the feature vectors, and the database of similarity is retrieved to match the most similar images, and then extracted from the database.
Gabor filters are a common choice for extracting texture features from images. A Gabor filter is a set of wavelets, each capable of capturing energy at a particular frequency and direction. The tunable scale and directional characteristics of Gabor filters make them play an important role in texture analysis. However, Gabor filters generally have a dc component and the response of the filter will depend on the grey values of the region. Therefore, a constant region with a high gray value will produce a high response, in practical cases, a high response will be produced only in the case of an edge or a corner, but not in a bright but constant region, that is, the existing solution has defects due to the existence of a dc component, such as a low accuracy and a long time consumption of the retrieval method.
Disclosure of Invention
The invention aims to provide an image texture retrieval method which is used for solving the problems of low accuracy, long time consumption and the like of the conventional retrieval algorithm.
In order to achieve the purpose, the invention provides the following technical scheme:
according to a first aspect, embodiments of the present invention provide an image texture retrieval method, in which a Log-Garbor filter-based image texture retrieval algorithm is provided, a direct current component of a conventional retrieval algorithm is eliminated, and when an image is observed on a logarithmic frequency scale, a response of the image is found to be gaussian, so that more information can be captured. Local texture description (LBP) is introduced, so that small and fine image details can be coded, the characteristic vector of the Maximum Edge Position information (LMEPOP) is obtained, and the accuracy of image retrieval is further improved. The method comprises the following steps:
1) converting a target image to be retrieved from a spatial domain into a frequency domain;
2) performing convolution operation on the transformed target image and a Log-Gabor filter;
3) performing inverse Fourier transform on the convolution operation result to obtain a Log-Gabor response image;
4) introducing local texture features (LBP) to obtain a maximum edge position LMEPOP feature vector of the Log-Gabor response image;
5) and retrieving similar images from an image library according to the extracted feature vectors, wherein the feature vectors of all the images are stored in the image library.
The foregoing description is only an overview of the technical solutions of the present invention, and in order to make the technical solutions of the present invention more clearly understood and to implement them in accordance with the contents of the description, the following detailed description is given with reference to the preferred embodiments of the present invention and the accompanying drawings.
Drawings
FIG. 1 is a flowchart of a method for retrieving image texture according to an embodiment of the present invention;
FIG. 2 is a Log-Gabor response image of a target image at different orientations and different scales according to an embodiment of the present invention;
fig. 3 is a method for calculating the features of the epop and mmopp in the 3 × 3 neighborhood of the gray-scale image according to an embodiment of the present invention;
fig. 4 is a block diagram of an image retrieval process according to an embodiment of the present invention.
Detailed Description
The technical solutions of the present invention will be described clearly and completely with reference to the accompanying drawings, and it should be understood that the described embodiments are some, but not all embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
In the description of the present invention, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc., indicate orientations or positional relationships based on the orientations or positional relationships shown in the drawings, and are only for convenience of description and simplicity of description, but do not indicate or imply that the device or element being referred to must have a particular orientation, be constructed and operated in a particular orientation, and thus, should not be construed as limiting the present invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and are not to be construed as indicating or implying relative importance.
In the description of the present invention, it should be noted that, unless otherwise explicitly specified or limited, the terms "mounted," "connected," and "connected" are to be construed broadly, e.g., as meaning either a fixed connection, a removable connection, or an integral connection; can be mechanically or electrically connected; they may be connected directly or indirectly through intervening media, or they may be interconnected between two elements. The specific meanings of the above terms in the present invention can be understood in specific cases to those skilled in the art.
In addition, the technical features involved in the different embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.
Referring to fig. 1, a flowchart of a method for retrieving an image texture according to an embodiment of the present application is shown, where as shown in fig. 1, the method includes:
the target image is converted from the spatial domain into the frequency domain using a Fast Fourier Transform (FFT).
102, performing convolution operation on the converted target image and a Log-Gabor filter;
the Log-Gabor function used by the Log-Gabor filter is defined as follows:
where (r, θ) represents coordinates in a polar coordinate system, f 0 Is the center frequency of the filter, which has a relationship f with the dimension n 0 ＝wav×scaleFactor n Wav is wavelength; theta 0 Is the azimuth angle of the minimum-scale filter; the scaleFactor is continuousScaling coefficients between filters; sigma r The scale bandwidth is determined; sigma θ The angular bandwidth is shown. In the present invention, four directions (0, pi/2, pi, 3 pi/2) and three scales (n is 0,1,2) are used, each scale generates a filter bank with 12 Log-Gabor filters, wherein wav is 3.0, scaleFactor is 3, and sigma is 3 r ＝0.65。
103, performing inverse Fourier transform on the convolution operation result to obtain a Log-Gabor response image;
the Log-Gabor response image is: i is m,n ＝|IFFT(FFT(I).*G m,n )|
I is the target image, representing performing a convolution operation in fourier space, G m,n Is the Log-Gabor filter.
Log-Gabor filter array G m,n Is the same size as the target image I to be filtered. For example, please refer to fig. 2, which shows the Log-Gabor response image at different directions and different scales of the target image.
104, acquiring an LMEPOP characteristic vector of the Log-Gabor response image through local texture characteristics (LBP);
in practical implementation, the step can be implemented as follows: obtaining a Local Maximum Edge Position actual Pattern (LMEPOP) feature vector of the Log-Gabor response image through Local texture features (LBP), wherein the Maximum Edge Position information is defined as symbol Maximum Edge Position actual Pattern (SMP) and amplitude Maximum Edge information (MMEPOP).
Optionally, the step of obtaining an LMEPOP feature vector includes:
firstly, calculating the difference value between the gray value of a central pixel of the Log-Gabor response image and a pixel in the field;
the LBP texture features are introduced to express the image feature vectors, and the LBP features have the remarkable advantages of gray scale invariance, rotation invariance and the like and are simple and easy to calculate. It assigns a code to each pixel of the image and thresholds the 3 x 3 neighborhood of each pixel by using the gray values. Firstly, calculating the difference between the gray value of the central pixel and the pixels in the 3 × 3 neighborhood thereof, wherein the calculation formula is as follows:
f(i)＝I(g i )-I(g c ),i＝1,2,...,8；
wherein, I (g) c ) Is the gray value of the central pixel, I (g) i ) Are the gray values of the eight neighborhood pixels.
Secondly, determining an optimal symbol SMEPOP characteristic and an optimal amplitude MMEPOP characteristic according to the difference. (ii) a
(1) Converting the difference value into a binary value;
the binary system obtained by conversion is:
the 8-bit binary number of the central pixel (usually converted into decimal number, i.e. LBP code, 256 kinds in total) can be obtained through the above steps, i.e. the LBP value of the central pixel point of the window is obtained, and the value is used to reflect the texture information of the area.
(2) The invention determines the symbol information index according to the binary difference value s And amplitude information index m 。
The LMEPEP operation is on the 3 × 3 neighborhood of the center pixel. Firstly, the gray values f (i) of the central pixel and eight neighbors thereof are calculated by the formula, then the values of f (i) are sequenced according to the signs and the amplitudes, and the obtained sign information index s And amplitude information index m The calculation formula is as follows:
index s ＝sort[f(1),f(2),...,f(8)]
index m ＝sort[|f(1)|,|f(2)|,...,|f(8)|]
(3) according to the symbol information index s And amplitude information index m Determining a dominant maximum SMEPEP feature and a MMEPEP feature.
In order to improve the identification capability, local LMEPOP is adopted to represent the LBP texture characteristics of the image, namely the edge position information with the first three greatest advantages is utilized. Based on the symbol information index s And amplitude information index m The three dominant maximum SMEPEP and MMEPEP characteristics are calculated by the following formulas. Fig. 3 illustrates a method of computing the SMEPOP and MMEPOP features within a 3 x 3 neighborhood of a grayscale image.
Thirdly, determining the LMEPOP characteristic vector according to the SMEPOP characteristic and the MMEPOP characteristic of the optimal symbol.
After computing the optimal SMEPEP and MMEPEP characteristics for each pixel, the entire image may be represented by a histogram using the optimal SMEPEP and MMEPEP. Since each of the optimal smepp and MMEPOP codes has a value between 0 and 501, there are 1004 image LMEPOP feature vectors that are finally obtained by connecting smepp and MMEPOP histograms.
And 105, retrieving similar images from an image library according to the extracted feature vectors, wherein the feature vectors of all the images are stored in the image library.
Optionally, the similarity between the extracted feature vector and the feature vector in the database may be calculated, and a similar image of the target image may be retrieved according to the calculated similarity.
Optionally, n images with high matching degree may be found from the database by measuring a chi-square distance between the feature vector of the target image and the feature vector of the image in the database, and the similarity calculation formula is defined as follows:
wherein, I m Representing the m-th image in the image database, I q On behalf of the query image(s),
a feature vector representing the query image,
representing the m-th picture in the image library, L f Representing the size of the feature vector. d (I) q ,I m ) Statistic-representative I q And I m Degree of match between, d (I) q ,I m ) Smaller values indicate higher matching levels.
Through the calculation, the image with the similarity ranking at the top n can be selected as the similar image.
It should be noted that the feature vectors of each image in the image library are similar to the feature vector extraction method described in step 101 to step 104, and this embodiment is not repeated herein. Fig. 4 is a retrieval flow diagram of the image retrieval method according to the embodiment.
In summary, the target image to be retrieved is converted from the spatial domain to the frequency domain; performing convolution operation on the transformed target image and a Log-Gabor filter; performing inverse Fourier transform on the convolution operation result to obtain a Log-Gabor response image; extracting a feature vector of the Log-Gabor response image; and retrieving similar images from an image library according to the extracted feature vectors, wherein the feature vectors of all the images are stored in the image library. The method solves the problems of low retrieval accuracy and long time consumption caused by the existence of the direct current component in the prior art, and achieves the effects of improving the retrieval accuracy and shortening the time consumption of retrieval.
Meanwhile, small and precise image details can be coded by introducing LBP, LMEPOP characteristic vectors are obtained, and accuracy of image retrieval is further improved.
The above-mentioned embodiments only express several embodiments of the present invention, and the description thereof is more specific and detailed, but not construed as limiting the scope of the invention. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the inventive concept, which falls within the scope of the present invention. Therefore, the protection scope of the present patent shall be subject to the appended claims.
Claims (9)
1. An image texture retrieval method, characterized in that the method comprises:
converting a target image to be retrieved from a spatial domain into a frequency domain;
performing convolution operation on the transformed target image and a Log-Gabor filter;
performing inverse Fourier transform on the convolution operation result to obtain a Log-Gabor response image;
acquiring a maximum edge position octal mode LMEPOP characteristic vector of the Log-Gabor response image through a local texture characteristic LBP;
retrieving similar images from an image library according to the extracted feature vectors, wherein the feature vectors of all the images are stored in the image library;
obtaining an LMEPOP feature vector of the Log-Gabor response image based on a local texture feature LBP, wherein the LMEPOP feature vector comprises the following steps: calculating the difference value between the gray value of the central pixel of the Log-Gabor response image and the pixels in the field;
determining an optimal symbol SMEPOP characteristic and an optimal amplitude MMEPOP characteristic according to the difference;
and determining an LMEPP characteristic vector according to the SMEPP characteristic of the optimal symbol and the MMEPP characteristic of the optimal amplitude.
2. The method of claim 1, wherein the defined Log-Gabor function:
where (r, θ) represents coordinates in a polar coordinate system, f 0 Is the center frequency of the filter, which has a relationship f with the dimension n 0 ＝wav×scaleFactor n Wav is wavelength; theta 0 Is the azimuth angle of the minimum-scale filter; scaleFactor is the scaling factor between successive filters; sigma r DeterminingThe scale bandwidth is reduced; sigma θ The angular bandwidth is represented; using four directions (0, pi/2, pi, 3 pi/2), three scales (n 0,1,2), each of which produces a filter bank with 12 Log-Gabor filters, wav 3.0, scaleFactor 3, sigma r ＝0.65。
3. The method of claim 2, wherein performing an inverse fourier transform on the convolution operation results to obtain a Log-Gabor response image comprises:
the Log-Gabor response image is: i is m,n ＝|IFFT(FFT(I).*G m,n )|；
I is the target image, representing performing a convolution operation in fourier space, G m,n Is the Log-Gabor filter.
4. The method of claim 1,
the difference is: f (I) ═ I (g) i )-I(g c ),i＝1,2,...,8
Wherein, I (g) c ) Is the gray value of the central pixel, I (g) i ) Are the gray values of the eight neighborhood pixels.
5. The method of claim 1, wherein determining an optimal sign SMEPEP signature and an optimal magnitude MMEPEP signature based on the difference comprises:
converting the difference value into a binary value;
determining symbol information index according to the binary difference value s And amplitude information index m ；
According to the symbol information index s And amplitude information index m Determining a dominant maximum SMEPEP feature and a MMEPEP feature.
6. The method of claim 5, wherein the binary value of the difference is:
7. the method of claim 5,
the symbol information index s Comprises the following steps: index s ＝sort[f(1),f(2),...,f(8)]
The amplitude information index m Comprises the following steps: index m ＝sort[|f(1)|,|f(2)|,...,|f(8)|](ii) a The sort function orders the values of f (i) by sign and magnitude, i ═ 1, 2.
8. The method of claim 5,
the optimal symbol SMEPEP is characterized in that:
the optimal amplitude MMEPEP is characterized in that:
9. the method of claim 1, wherein determining the LMEPOP feature vector based on the optimal symbol SMEPOP feature and an optimal magnitude MMEPOP feature comprises:
the whole image is represented by a histogram using the optimal symbol SMEPEP feature and the optimal amplitude MMEPEP feature;
and connecting the SMEPEP feature of the optimal symbol and the MMEPEP feature histogram of the optimal amplitude to obtain an LMEPEP feature vector of the image.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011603191.2A CN112667842B (en) | 2020-12-29 | 2020-12-29 | Image texture retrieval method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011603191.2A CN112667842B (en) | 2020-12-29 | 2020-12-29 | Image texture retrieval method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112667842A CN112667842A (en) | 2021-04-16 |
| CN112667842B true CN112667842B (en) | 2022-08-02 |
Family
ID=75410634
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011603191.2A Active CN112667842B (en) | 2020-12-29 | 2020-12-29 | Image texture retrieval method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112667842B (en) |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107958073A (en) * | 2017-12-07 | 2018-04-24 | 电子科技大学 | A kind of Color Image Retrieval based on particle swarm optimization algorithm optimization |
| CN107977664A (en) * | 2017-12-08 | 2018-05-01 | 重庆大学 | A kind of road vanishing Point Detection Method method based on single image |
| CN108596250A (en) * | 2018-04-24 | 2018-09-28 | 深圳大学 | Characteristics of image coding method, terminal device and computer readable storage medium |
-
2020
- 2020-12-29 CN CN202011603191.2A patent/CN112667842B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107958073A (en) * | 2017-12-07 | 2018-04-24 | 电子科技大学 | A kind of Color Image Retrieval based on particle swarm optimization algorithm optimization |
| CN107977664A (en) * | 2017-12-08 | 2018-05-01 | 重庆大学 | A kind of road vanishing Point Detection Method method based on single image |
| CN108596250A (en) * | 2018-04-24 | 2018-09-28 | 深圳大学 | Characteristics of image coding method, terminal device and computer readable storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112667842A (en) | 2021-04-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Bagri et al. | A comparative study on feature extraction using texture and shape for content based image retrieval | |
| Kandaswamy et al. | Efficient texture analysis of SAR imagery | |
| Parisi-Baradad et al. | Otolith shape contour analysis using affine transformation invariant wavelet transforms and curvature scale space representation | |
| US8718380B2 (en) | Representing object shapes using radial basis function support vector machine classification | |
| KR100836740B1 (en) | Video data processing method and system thereof | |
| Alazawi et al. | Texture features extraction based on GLCM for face retrieval system | |
| CN110717497B (en) | Image similarity matching method, device and computer readable storage medium | |
| US10866984B2 (en) | Sketch-based image searching system using cell-orientation histograms and outline extraction based on medium-level features | |
| CN111400528B (en) | Image compression method, device, server and storage medium | |
| CN104850822B (en) | Leaf identification method under simple background based on multi-feature fusion | |
| CN111507324A (en) | Card frame identification method, device, equipment and computer storage medium | |
| CN114255468A (en) | Handwriting recognition method and related equipment thereof | |
| Çevik et al. | A novel high-performance holistic descriptor for face retrieval | |
| CN110555406B (en) | Video moving target identification method based on Haar-like characteristics and CNN matching | |
| CN110795995B (en) | Data processing method, device and computer readable storage medium | |
| CN112667842B (en) | Image texture retrieval method | |
| Nam et al. | A shape-based retrieval scheme for leaf images | |
| Guo et al. | Fault diagnosis of power equipment based on infrared image analysis | |
| CN108334884B (en) | Handwritten document retrieval method based on machine learning | |
| Mumar | Image retrieval using SURF features | |
| Nam et al. | Elis: An efficient leaf image retrieval system | |
| Wu et al. | A hierarchical feature search method for wine label image recognition | |
| Lakshmi et al. | Plant leaf image detection method using a midpoint circle algorithm for shape-based feature extraction | |
| Diaz-Escobar et al. | Natural scene text detection and recognition with a three-stage local phase-based algorithm | |
| Ouslimani et al. | Directional rank coding for multi-scale texture classification |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |