Study On Analysis And Classification Of Texture Images

The analysis and classification of texture images is researched in this dissertation, including the feature extraction of texture images, texture pattern cluster and segmentation etc. For this purpose, a wavelet packet decomposition and texture feature extraction method has been studied, and the BP neural network model has been used in classification stage. In order to handle the problem of detection of texture defects, a new concept of different adaptive wavelet bases has been proposed using genetic algorithms, the texture defects can be detected and located properly. In this dissertation, the conventional Gabor filter has been researched, and a multiresolution binomial distribution filter has been presented, the precision of texture segmentation edge has been improved. This dissertation also presents a nature texture classification method based on fractal dimension using self-similar texture characterization. For this purpose, the fractal dimensions are estimated with the Modified Box-Counting Method. The result has as good performance as that of other methods. A modified texture segmentation algorithm based on multiresolution model is proposed in this dissertation. The method results in accurate segmentations and requires significantly less computation. In this dissertation, an evolutionary programming-based sampling frequency-sensitive network is proposed for pattern clustering. The proposed method has an advantage of that the optimal solution of architecture and parameters can be got simultaneously. So the classification network canget the optimal number of clusters and the optimal vector quantization.The achievements of the dissertation include:A wavelet packets model reflecting multichannel filtering method is used for texture classification. Pyramid wavelet decomposition coefficients are not used as the feature inputs of this model. The tree structure wavelet decomposition (wavelet packets ) method is used to get energy features at the spatial frequency domain of the texture. It is because that using a single scale analysis has been considered having low texture classification ability. So using the multiresolution wavelet transform can overcome this disadvantage. But Mallat pyramid wavelet decomposition algorithm decomposes texture image iteratively at the low spatial frequency channel. A large amount of nature texture images are considered having quasi-periodic modes, so the Mallat pyramid decomposition method can not get the enough message for texture classification. The energy information of middle spatial frequency can be get using tree structure wavelet decomposition. Finally the BP neural network model has been used in classification stage.There are many textures such as woven fabrics having repeating textrons. In order to handle the textural characteristics of images with defects, this dissertation proposes a new method based on two-dimensional wavelet transform. In the method, a new concept of different adaptive wavelet bases is used to match the texture pattern. The 2D-wavelet transform has two different adaptive orthonormal wavelet bases for rows and columns which differ from Daubechies wavelet bases. The orthonormal wavelet bases for rows and columns are generated by genetic algorithm. The experimental result demonstrate the ability of the different adaptive wavelet bases to characterize the textureand locate the defects in the texture.This dissertation proposed a multiresolution binomial distribution filter and presents a performance analysis and comparison between the binomial distribution filter and Gabor filter in the space domain and frequency domain. When the scale space constants of the two filters are large enough, the characteristics and performances are almost the same. But when the scale space constants are smaller, the conclusion can be made that the performances of the binomial distribution filter are better than that of Gabor filter. The interpolation and iterated algorithm can be used in the calculation of the binomial filter. The texture image classification expe...