Original link: http: //www.51jrft.com/dzxx/dzkxyjs/561.html
Having different characteristics depending on the texture of the seabed type, as texture features based on the feature classification. Use GLCM texture feature extraction method to obtain the desired texture characteristics test, and finally completed the classification using support vector machines to obtain classification accuracy is preferable. Demonstrated using the method of texture feature extraction and GLCM SVM seabed side scan sonar image classification is possible.
Keywords: side-scan sonar image; wavelet transform; texture feature; support vector machine; image classification HM000069
3.4 texture feature extraction simulation and implemented
in MATLAB , the gray level co implemented by calling a predetermined function, the following function creates a brief graycomatrix GLCM.
Call graycomatrix GLCM function can create an image, the basic call is:
GLCM = graycomatrix (the I);
call the function can be created by the image gray level co GLCM I. Calculating the frequency of the pixel having a gray level of i and j to the adjacent gray level in the horizontal direction appears. Each element described glcm the number of occurrences of the horizontal adjacent pixels. If the gray level is the L dimension is glcm L * L. In MATLAB, 8-bit grayscale.
In order to meet different needs, may be provided more parameters when invoked, calls the method is:
glcms = graycomatrix (the I, the param1, val1 is, param2, val2 ...)
can return the parameters set in accordance with the demand or a plurality gray co-occurrence matrix. The following brief description of the parameters:
GrayLimits: gray boundaries, is a binary vector [low, high]. Gray scale value less than or equal to 1 corresponds to low, greater than the corresponding gray level is equal to high. If [], using the co-occurrence matrix representing the minimum and maximum grayscale image gradation value limit.
NumLevels: specified as an integer. DESCRIPTION gray scaled number of gray levels in the image I. Determines the number of gray levels glcm size. By default, the digital image is 8, the binary image is 2.
Offset: p rows and two columns of the integer matrix described distance between the pixel of interest and adjacent pixels.
Extracting feature values graycoprops in MATLAB, the function is called based on the use of the function graycomatrix. Call as follows:
stats = graycoprops (GLCM, Properties);
wherein an effective GLCM GLCM. If glcm co-occurrence matrix is a matrix of gray, each matrix comprising the stats GLCM static property, the default parameters of properties equivalent to all, calculated values of four properties.
Graycoprops GLCM calculated using normalized attribute, attribute parameters are as follows:
Contrast: Contrast. Back in brightness between the pixels of the entire image and its neighboring pixels. Consistent image gray value, the contrast is equal to zero.
Correlation: correlation. Back to measure the entire image pixel and its neighboring pixels is related to how. In the range [-1,1]. Consistent image gray value, the correlation is NaN.
Energy: energy. Returns the element of the square and GLCM. In the range [0,1]. Consistent image gray value, energy is equal to 1.
Homogemeity: homogeneity. Returns metrics GLCM distribution of elements close to the diagonal extent. In the range [0,1]. Homogeneity is a diagonal matrix.
The following code object image and generating GLCM feature value extracting:
the IN = imread ( 'chonggou.bmp');
G = 256;
[R & lt, C] = size (the IN);
P1 = zeros (G );% calculated co-occurrence matrix
for M =. 1: R & lt
for N =. 1: (. 1-C)
P1 (the IN (M, N) +. 1, the IN (M, N +. 1) + 1'd) = P1 (the IN (M , N) +. 1, the IN (M, N +. 1) + 1'd) + 1'd;
P1 (the IN (M, N +. 1) +. 1, the IN (M, N) + 1'd) = P1 (the IN (M, N + 1'd) +. 1, the IN (M, N) + 1'd) + 1'd;
End
End
P2 = Double (P1) / (M * N);% normalized
obtained by running the above code result P2, then P2 using feature extraction, specific code as follows:
% calculated angular second moment F1
F1 = P2 ^ 2;.
F1 = SUM (F1 (:))
% calculated correlation F2
Color = 0: (. 1-G);
M1 = SUM (P2) Color * (:);
M2 = M1;
PP = SUM (P2);
V1 = ((M1-Color) ^ 2) * PP (:);.
V2 = V1;
[m, n-] = meshgrid now (Color, Color);
PP = m * n-* P2;..
F2 = (SUM (PP (:)) - M1 * M2) / (V1 + EPS) * (V2 + EPS)
% calculated moisture F3
; PP = (. * log2 P2 (P2 + eps))
F3 = -sum (PP (:))
Abstract the I
the ABSTRACT II
first chapter 1
background and significance of topic 1.1 1
research status at home and abroad 1.2 1
1.3 2 side scan sonar imaging features
1.4 topic 3 work
Chapter 4 submarine sonar image preprocessing
2.1 Introduction 4
Submarine sonar image enhancement 4 2.2
2.2.1 histogram equalization 4
2.2.2 image sharpening 7
2.2.3 smoothed image 9
2.3 wavelet transform 12
2.3.1 13 D wavelet transform defined
2.3.2 dimensional Wavelet transform and denoising 14
2.4 15 Summary
Chapter seabed texture feature extraction 16
3.1 Introduction 16
3.2 Overview of image texture feature extraction method 16
3.3 18 extracts the image texture features based on the seabed GLCM
texture feature 3.3.1 GLCM principle 18
3.3.2 gray level co-occurrence matrix texture feature value introduced 19
simulation 3.4 texture feature extraction and Implementation 21
3.5 Summary 24
Chapter 26 seabed image classification based on SVM
4.1 Introduction 26
4.2 Introduction 26 SVM
4.2. 1 the general characteristics of the SVM 26 is
4.2.2 separable linear and nonlinear SVM separable description 27
4.2.3 SVM kernel 29
4.3 seabed image classification Kit libsvm profile 29
during installation and 4.3.1 libsvm Notes 30
4.4 Based on the seabed sediment texture feature SVM classification general process 31
4.4.1 sample database build 31
4.4.2 SVM-based image classification submarine simulation and the 34
41 4.5 Summary
Chapter Summary and Outlook 42
5.1 summarizes 42
5.2 Outlook 42
References 44
Acknowledgments 46
original link: http: //www.51jrft.com/dzxx/dzkxyjs/561.html