Study On Some Methods Of Image Textural Features Representation And Their Applications

Along with the rapid development of science and technology, the images become a more important data carrier in the field of information transmission. The effective image representation is the important foundation for further image processing. Most of images obtained from various observation systems are texture-based image. They can be analyzed by extracting macroscopical information. Because of the complexity of texture, the precise definition of texture image has not yet been given, which brings considerable difficulty for texture analysis. Generally, texture refers to the local repetitive patterns and their arrange rules. The texture features are the quantization values of gray-level changes within the region. There are three basin texture features, including periodicity, directionality and randomness. Periodicity and directionality are high-level texture features, which can be used to guide the perceived perception for texture images. This thesis studies the representation theory of texture features. Textile image processing is the application background. The following are the main contribution of this thesis.1. Periodicity is an important texture feature. This thesis outlines periodicity extraction methods and analyzes the principle of Fourier transform for extracting periodicity. It also presents the measurement of texture periodic feature. The woven fabric in the textile industry is a typical periodic image. This thesis presents a parameter to depict the periodicity of fabric images. Then the parameter is applied to the automatic identification of the fabric structure, which achieves recognition plain cloth, denim cloth and urgency denim cloth. The precise float can be calculated too. The algorithm reaches satisfactory results and overcomes the shortcoming of the traditional method, which need to damage the fabric sample. It also improves the efficiency of the past methods based on digital image processing. At the same time, for the inevitable skewness in the process of getting fabric images, this thesis proposes a rapid fabric rectifying method based on Fourier transform as one of the preprocessing steps. This method can also be extended to the text skew correction. Combining our method and the denim colors computing, as well as the calculation of counts et al., a denim management and analysis system is developed, which will be soon used in the process of production and management by a textile factory in Hong Kong.2. Directionality is an important visual feature of texture too. This thesis studies the past methods. It presents a robust image texture directional measurement method. Gabor filters is much more suitable for simulating human vision system and can achieve the localization both in the time domain and frequency domain. The principle of enhancing directional features is analyzed. This thesis presents two extracting directional methods in the time domain and frequency domain respectively. One method designs a cost function to select the optimal Gabor filter in the time domain. According to the characteristics of target, the other method calculates the parameters for designing the optimal Gabor filter in the frequency domain. The parameters estimation method is also presented. Then the new methods are applied to extract slub, which is the directional feature of fabric images. Instead of the past extracting method, in which the slub is regarded as fabric defects, slub is regarded as a texture feature. A more suitable Gabor filter is designed. The above methods have been compared with the existing method proposed by G. Pang. The experimental results demonstrate the feasibility and validity of our methods. Our method is also applied to local finger image enhancement, which gets satisfactory results.3. We study the representation methods of abrupt features in the periodic image, which wavelet transform is one of the most important methods. The abrupt feature extraction based on wavelet transform enhances the regional features of abruption, and then segment abrupt region through thresholding. The important step is the selecting or designing of wavelet basis. For the existing wavelet basis can not reflect the considering image features adaptively, the adaptive wavelet construction model for extracting abrupt features is proposed. The solving method and the experimental results are also demonstrated. The solution to this model can be well described the periodic background texture. It achieves the purpose of enhancing abrupt features. Our method overcomes the shortcoming of the traditional non-adaptive wavelet methods. The feasibility and effectiveness are shown from the perspective of experimental verification.