Research And System Implementation Of Image Annotation Based On Bayesian Method

Image annotation is a process that annotate a image with semantic labels, to express the content of the image. With the development of Internet, imaging and storage technology, huge amounts of image data is producted anytime anywhere, how to carry on the classification and retrieval of these visual information, has become a hot topic. Image annotation is the basic premise for image retrieval, involving all aspects of image processing. In the development of computer science, its importance is self-evident. Since its birth, each computer expert pay attention to it, in order to make significant progress in this field, laying foundation of the research and development in the field of image.Due to the existence of the semantic gap between high-level semantic information and low-level visual features, the accuracy of the existing image annotation models is often not satisfactory, and hard to work well on the big data. Therefore, based on the study of existing methods, this paper try and explore from multiple perspectives, to work out how to solve the problem of semantic gap, how to improve the effect of image annotation, how to improve the efficiency of the labeling process, the details is as follows:①This paper mainly studies the annotation model based on Bayesian method, introduces its basic framework, analyses the main methods, including parameter method and nonparametric method, and compares the advantages and disadvantages with annotation algorithms.②In this paper, an efficient annotation algorithm is proposed, using the annotation model framework based on Bayesian method. In this paper, the optimization methods are used in each part, to improve the effect and efficiency of image annotation. During image segmentation, fixed division and cluster are employed, mixing together segmentation and training; in feature extraction, domain color descriptor and Gabor filter texture are chosen and combined; adaptive divisive hierarchical k means clustering is used; the word-to-word correlations is considered in semantic annotation. Then this paper makes the detailed describe about each part of the method. The three groups of experiments are designed, to verify performance of the method on different features, validity and applicability. The method achieves good efficiency and effectiveness, reaching the expected goal.③According to the annotation method proposed in this paper, an automatic image annotation system is designed and implemented. This article first makes the system requirement analysis, then introduces the system’s overall architecture, detailed design of each function module and database, finally implements annotation processing algorithms by MATLAB. The automatic annotation system is integrated, displayed and explained. This system can be used to accomplish the actual image annotating work, also to make a comparative analysis of annotation algorithms.