Image Texture Feature Extraction And Image Classification System Research And Implementation

With the fast development and wildly application of the Internete, digital multimedia and intelligent information processing technology, more and more digital images came out. Facing to so many of the image information, searching and finding the images with right content becomes harder and harder. The need to find an efficient way to manage and retrievel images is eager. The content-based image retrieval technology has attracted much interest from several study fields, such as image processing, pattern recognition, machine learning, etc. The image information description/extraction and the image classification are two important processes in the content-based image retrieval. This article studied the image classification method based on texture feature extraction. Using an improved Grey Level Co-occurrence Matrix (GLCM) algorithm with the multi-class SupportVector Machines (m-SVMs) technology, the image classification system achieved good classification results.First, aiming at the problem of image feature extraction, this article researched many kinds of methods about describing and extracting image texture feature, which is one if the low level content of image. Taking texture characters and the practical application of the content-based image retrieval technology into account, this article chose the GLCM algorithm as the object of study. The research fond the GLCM algorithm has a lot of redundant calculation, and need a big memory space. Thinking about the defect, this article studied the other improved algorithm based on GLCM algorithm, and put forward an improved algorithm which is called Statistics Grey Level Co-occurrence Matrix and Link algorithm. It is easy to find that the new algorithm needs less memory space than GLCM algorithm.Through the experiment, comparing the comuptition time of GLCM algorithm and Statistics Grey Level Co-occurrence Matrix and Link algorithm, it is proved that the later one will use less time to compute the feature values.Secondly, aiming at the problem of image classification, this article researched many kinds of methods. Because of the SVM has a strong mathematics theoretical principle and a good classification capacity, this article chose the SVM technology as the object of study. With the Statistics theoretical principle and classification principle of SVM, this article studied three factor of a SVM model, kernel function, training algorithm and multi-class classification algorithm. Thinking about the image classification characters, this article put forward the m-SVMs model fitting for image classification. With the research upwards, this article designed an image classification system based on texture feature extraction. Achieving the Statistics Grey Level Co-occurrence Matrix and Link algorithm and m-SVMs model, the system owned the capability to extraction texture feature, training and classification the images. The experiment proved that the image classification system based on texture feature extraction achieves good image classification precision.Finally, this article talked about the further of the content-based image retrival technology from several fields, including feature extraction, supervised learning and classification feedback.