Research On Algorithms Based On Finite Mixture Model For Automatic Image Annotation

As the development of the technology of network, there is an increasing amount of information on the Internet in today's society. Text-based image retrieval and content-based image retrieval as their shortcomings, certainly can not meet people's needs, thus promoting the automatic image annotation technology.The essence of automatic image annotation is to obtain the semantic key words of images from visual features and to support the semantic level search, then image retrieval can be transformed into text retrieval, which is fairly mature. To a certain extent, it can solve the semantic gap existing in content-based image retrieval.Automatic image annotation is based on image visual features. It involves image segmentation and image feature extraction. Thus, image segmentation algorithm and image visual feature extraction method are briefly introduced at the beginning of this paper. On this basis, we expand the discussion of the next level.This paper focuses on finite mixture model based automatic image annotation. Firstly, we introduce the theory of finite mixture model and EM algorithm for maximum likelihood estimation. Then, In accordance with Gaussian mixture model and t-mixture model, we discuss these two methods one by one and study them for automatic image annotation. During the process of using Gaussian mixture model to solve automatic image annotation, we investigate the estimation of parameters using EM algorithm. And then, taking into account the specific process of automatic image annotation, we built the automatic image annotation model based on Gaussian mixture model.During the process of using t-mixture model to solve automatic image annotation, we also investigate the estimation of parameters using EM algorithm. Meanwhile, due to the shortcoming of the EM algorithm, we construct SMEM algorithm for them. At last, taking into account the specific process of automatic image annotation, we built the automatic image annotation model based on t-mixture model.