The Research Of Image Semantic Extraction And Analysis Of Multi Feature Fusion

Along with the rapid development of Internet in recent years, the image data is growing bigger and bigger. Since the data is myriads of changes visually and different semantic dispersion, users are more and more desiredly need a fast and convenient retrieval way for large image database, then the image semantic extraction and analysis technique emerges as the times require. Image semantic extraction is the precondition and key step of semantic analysis and image retrieval, but the low-level features of image is difficult to automatically derive high-level semantic.So solving the "semantic gap" has always been the hot topic of image semantic field. This paper focuses on the image semantic extraction field studying the image semantic extraction and analysis of multi feature fusion based on technique of support vector machine.Firstly,the paper describes several common image features including color histogram, gray level co-occurrence matrix, shape contour feature extraction methods. It also introduces the basic principle of support vector machine and its the kernel function model.It provides a theoretical basis for follow-up research firm.Since the accuracy rate of traditional method of multi feature fusion of image semantic extraction is not high,we also use the serial and parallel feature fusion in this paper proposing the idea of M feature extraction method and combining their advantages to study the existing multi feature based on the kernel method. We get the M feature using fusion method from image color, texture and shape features extracted. Based on the above ideas, this paper designed the experiments, provided the experimental results, and the experimental comparison, the experiment proves that this algorithm has higher accuracy rate to extract image semantic.Finally, since the SVM classifier has the choice of kernel function and its kernel functions including radial basis kernel function, the linear kernel function, polynomial kernel function and Sigmoid kernel function. We take the radial basis kernel function, and integrate the K-Means cluster analysis method into SVM classifier.This paper also analyzes the shortcomings of this approach, and make the relevant improvement. The improved K-SVM is the low-level feature to high-level semantic mapping is the key to the implementation of the results. We also design the experiments, providing the experimental results, and the experimental comparison, the experiment proves that this algorithm to extract semantic accuracy.