Research On Texture Feature Extraction And Classification

Texture is an important attribute in the image,while provides substantial information for the recognition and interpretation of this image.The early methods analyze the texture at a single scale.Wavelet analysis is a new multiresolution analysis tool.It is appropriates for texture image processing because of its spatial-frequency localization performance and diversity.Support vector machine(SVM) is a general learning machine,and is a method of statistical learning theory,its main idea is structural risk minimization.It is widely used these years beeause of its perfect performance.This thesis studies texture feature extraction and classification based on support vector machines.The main contents and contributions are as follows:(1) An texture classification approach based on Stationary wavelet transform(SWT) and the gray-level co-occurrence matrix is proposed.Firstly,the texture images to be classified are translated via SWT.Then,the co-occurrence features are computed for different angles and averaged ,of original images, approximation of 1-level decomposed texture and averaged detail sub-bands.Finally,the support vector machines are used to the texture classification.The experiment results show the proposed method can get more results for texture classification.(2) As the scale and rotation invariant texture image classification is concerned,an classification approach based on log-polar transform and stationary wavelet transform is realized.Firstly,the texture images is translated via log-polar transform to convert the rotat- ion to translation .Then,texture feature is computed by SWT which is shift invariant to the translation.Finally,the support vector machines are used to the texture classification.(3) A directional multiresolution approach was proposed for texture classification based on a modified contourlet transform, It is implemented by the stationary wavelet and nonsubsampled directional filter banks (SWT-NSDFB).Then,take the energy measures of all the subbands including one lowpass band and every directional subband in each level as the texture feature.Finally,the support vector machines are used to the texture classification.