Content-based image retrieved with MPEG-7

Abstract

The major objective of this project is to build a content-based image retrieval system with use of the MPEG-7 technology. This system also has a very user friendly web-based interface. The main function of this system is to search similar pictures in a large image database by using images features descriptors such as color and texture, which defined in the MPEG-7 standard. Features of each picture in the image database are first extracted and presented in a compact form by extensible XML language, which could effectively increase the searching speed. In this project, a novel color searching algorithm called Key Color Descriptor was also developed and implemented in the system besides thebasic content descriptors defined in MPEG-7. This new searching method is implemented according to the MPEG-7 specification; it can match similar color images by means of color quantization and palette merging. Experimental results show that the proposed KCH descriptor achieves more precise retrieval result than the MPEG-7 Dominant Color descriptor. Moreover, we revise the similarity measure of Dominant Color Descriptor into MPHSM. Experimental results show that the proposed method achieves higher precision and recall rate than original one. A novel approach of relevant feedback was also implemented in this system for improving the retrieval accuracy.