Summary: | 碩士 === 義守大學 === 資訊工程學系 === 102 === In recent years, multimedia applications have quickly grown with the new technologies and innovations. Since the amounts of multimedia data are also increasing rapidly, the management and classification for multimedia data are becoming more and more important.
In this thesis, we first partition images to many blocks for extracting image features. The size of each block is 4 × 4 pixel. Based on the block content, we classify the blocks to two types of visual words: macro and micro visual words. After training procedure, we can derive macro and micro visual vocabulary, which are used to construct probability models for classification.
Experimental results indicate that the probability models support effective performance for classification. The optimum weighting values of macro and micro are automatically adjusted. Experiment results show that the proposed method can effectively determine the proper ratio.
|