Image Classification Based On Segmentation Comprehensive Feature Weighted By Genetic Algorithm

In recent years, with the development of multimedia, digital image is growing at an explosive pace, more and more information exist in the form of an image. Especially with the development of smart phones and mobile Internet terminals, so that the image and video have become the main form of multimedia. In the huge amount of image information, it is difficult to find the need information quickly and accurately. As an effective method of image retrieval, image classification has increasingly become a hot topic. There are many techniques of image classification, but there is not a general classification method applicable to all the image data sets. Therefore, how to improve the performance of image classification method is an important research topic in image research area.Content-based image classification method avoids time-consuming and inaccurate subjective judgment of artificial mark. But there are still some problems, for instance, the underlying characteristics of image expression of image high-level semantic features is prone to semantic gap. For the semantic gap, we propose a new classification technology based on segmented integrated features which are weighted by an improved genetic algorithm(IGA). First of all, with the image block method proposed in this paper, the image is divided into five parts, four characteristics of each part are extracted and integrated for the features of the entire image; Then the improved genetic algorithm is used to optimize the weights of different characteristics, which leads the composition of optimal synthesis features to reduce the semantic gap and well express the image information; Finally, support vector machine(SVM) is used to classify image.In this paper, the main work includes:(1)By using the complementary between global features and local features, this paper proposes a method of features fusion which bases on the color accumulate histogram, color moment, symbiotic torque and SIFT features.(2)The method based on center which make full use of the spatial distribution information of image. Then this method was compared with methods of literature, and the experiments show that this method is more effective to express the information of the image.(3)For the drawbacks of slow convergent speed and getting local optimum solution while using the genetic algorithm(GA), an improved GA is proposed and some simulation experiments are made based on three test functions.(4)According to important degree of different characteristics of expression image, the improved genetic algorithm for feature weighting is proposed, using the good search ability of IGA to find the optimal weights.(5)Support vector machine classification experiments are given for different feature extraction methods such as single feature,integrated features, block comprehensive features, and weighted comprehensive features,and the experimented results are analyzed.