Research And Application Of Image Retrieval Technology Based On Improved EM Algorithm And SVM With Mixtures Of Kernels

With the rapid development of multimedia and internet, people have paid more attention to image information. So it's significant to organize, manage, and use the image-based multimedia information. But there are still lots of defects in image processing efficiency and performance of relevance feedback of the content-based image retrieval recently. Much research is done in that field, and gets some achievements in this paper.Firstly, the research of content-based image retrieval systems and the key techniques of them are analyzed. After the introduction of image features, similarity calculation, feature extraction and relevance feedback, according to the retrieval requirements of the images, the content-based image retrieval based on the improved EM algorithm and SVM relevance feedback using mixtures of kernels is designed. This paper presents an integrated image retrieval method which uses texture features and shape features, and shape features are descriped by two representation methods of invariant moments and Fourier descriptors which can describe the shape features from different angles. For weakening the negative impact of the image background areas, this paper introduces a method of region-based color image segmentation which uses EM algorithm to get the main region. As well as the traditional EM algorithm was improved, the complexity of the algorithm is greatly reduced. In relevance feedback, in order to further improve the system's learning ability, this paper introduces mixed-kernels functions and genetic algorithms, and presents a kind of SVM which uses genetic algorithms to optimize the choosing of the kernel parameter, penalty factor and kernel coefficient, so improve the system's self-adaptive. The improved SVM relevance feedback algorithm makes image retrieval system have a higher adaptability and robustness which can meet the different types image searching under different conditions.Finally, this paper presents the implementation of an image retrieval system based on improved EM algorithm and SVM with mixtures of kernels. We introduce the application into the cutler image retrieval of office automation system in certain aircraft enterprise initially and achieve a series of good retrieval results.