Researches On Multi-Features Image Classification Approach Based On Two-Stage Segmentation

With the rapid development of computers, communications and multimedia, especially with the emergence of the Internet and its popularity, the growth of the image data extends in an explosive way.Thus, the traditional keyword based classification methods can not handle the massive data, and they are impossible to meet the need of individual. As a consequence, the based content classification techniques come into being.The Classification of Image is a challenging issue, which will contribute much to medical diagnosis, crime detection, GIS, et al. However, there is not an unified frame which can significantly improve the accuracy of all types of image classification approaches. In order to improve the classification results effectively, this thesis proposes several improvements in some key techniques and algorithms in image classification.(1)The effective image segmentation is important for image classification. A two-stage segmentation algorithm for color image presented. In other words, the image is firstly separated by SOM and then segmented by the K-means algorithm. Comparing with the traditional image segmentation approaches only based on SOM or SBN, the proposed approach can obtain better segmentation results.(2)Several features are applied for image classification synthetically. As the extraction of the image feature is a key technology in classification process and whether the image features can be formulated or not will directly affect the results of image classification. Because the single feature can not describe the image effectively and represent the image content well, the comprehensive features of the image such as the color, texture and shape are employed in image classification in this thesis. That is to say, totally 14 feature vectors are used to represent the image content. Some experiments are accomplished and experiment results are provided. (3)Support vector machine (SVM) learning approach is used for image classification. The advantage of the Support vector machine is that it needs not prior knowledge about specific issues, and it can solve the problem of small sample better. Two types of image, the simple images and complex images, are classified by SVM.The research results demonstrate that proposed technique and approach can improve the precision of the image classification efficiently. Thus, making use of these results, we will service in the image process domain well.