Study On Image Feature Extraction Based On MPEG-7

Image feature extraction is one of key techniques in content-based image retrieval research field. Feature extraction techniques which based on MPEG-7 standard can enhance the interoperability of content-based image retrieval systems. Two methods of image feature extraction which based on MPEG-7 standard were proposed in this thesis. The contributions of this thesis are list as follows:(1) Based on the analysis about scalable color descriptor, and considering HSV color model characteristic and human vision feature, a fuzzy color descriptor is proposed. According to human vision feature, the transition between two hues is a process that one becomes weaker and another becomes stronger. A fuzzy quantization strategy is proposed to linear simulate the transition between two hues. The experimental results show that performance of fuzzy color descriptor is better than that of scalable color descriptor.(2) A method of dominant color feature extracting is proposed. Dominant color is a compact and efficient descriptor which employs representative colors to characterize the color information in the interesting region of an image. Region growing algorithm is used to extract dominant color of images. The feature extracting process is described in detail in this thesis. Proposed method is compared with merged histogram method based algorithm. The experimental results show the proposed method get a better result.(3) Fuzzy color descriptor and dominant color descriptor are combined to apply in the content-based Web image retrieval engine prototype system which developed by our team. The experimental results show that the effects of combined color features which are composed of fuzzy color descriptor and dominant color descriptor are better than that of single color descriptor.