Image Classification Based On HSV And Texture Features

With the explosive growth of the images, how to use image features for effective classification and annotation is a hot topic in image classification. In order to allow users to efficiently find images from massive data, effective image classification and annotation is extremely important.In image classification we often fully explore the underlying characteristics of the images, which effectively express the image features for image classification. In order to improve the classification performance of the image classification technology based on content, image features such as the color, texture, shape, spatial information, and so on will be studied in this paper. In addition, the LBP operator and block strategy have been improved. These improvements are as follows:(1) A classification algorithm based on weighted HSV block is proposed. Although at present some block weighted algorithm can improve the classification performance in some certain extent, some block methods are complex or weight setting is unreasonable. So in this paper the image area first is divided into three blocks according to the level of interest:the center of the image is in focus, and the center of the image in circular region is given great weight; secondly, an elliptical region is set and given more weight in the circular area outside to ensure the second-importance of such information; the rest area is the edge region, and the importance in the image is relatively weak, thus it is given less weight.(2) The improved LBP operator is put forward. Although the traditional LBP descriptors or some improved LBP descriptors have certain anti-noise performance, the anti-noise ability decreased dramatically when there is the integration of large amount of noise. In order to overcome this defect, an improved LBP operator is proposed in this paper:To determine the threshold, the original 3*3 region is expanded to 5*5 region, which is conducive to reduce noise effects on average. Experiments show that the anti-noise performance of the improved LBP operator can get bigger promotion.(3) Classification algorithm based on multiple features is proposed. Although the average accuracy rate of classification performance for the single feature is good, the classification performance for the specific images of different categories is not uniform. The HSV of block weighted, improved LBP operator and GLCM based on multiple features will be combined. At the same time, in order to solve the problem of high dimension of extracted HSV and LBP, the dimensions of HSV and LBP respectively reduced by means of the PCA. Then GLCM is extracted from 4 irrelevant features in 4 directions. Finally, HSV, LBP and GLCM are combined. Experiments show that the multi-feature fusion classification method can obtain better classification performance.