Research On Image Retrieval Techniques Based On Content And MPEG-7

With the development of the multimedia and computer network technology, the application of the image is more and more extensive; Content-based image retrieval (CBIR) technology becomes one of the most active fields in multimedia retrieval. CBIR is an image retrieval technology, which synthesizes various visual features in digital image, such as color, textual, and shapes features. However, due to the complexity of the information, the lack of standard for the description of image content induces low interoperability among CBIR systems. To meet this challenge, MPEG-7 was developed as a solution to the normative description of audio-visual content. MPEG-7 relates the description with its content, and enables the effective and efficient query and access to multimedia content.This thesis focuses on the bottom basic features offered by MPEG-7, and combining MPEG-7 color, texture features and descriptor of the image, and the goal is to develop an image retrieval prototype system.Section I studies CBIR. By introducing the background of CBIR, this thesis gives several image retrieval methods based on color, texture and shape features.Section II introduces the MPEG-7 standard with the main elements, parts and the scope of the MPEG-7 standard, etc. This section emphasizes on the basic structures and descriptors that covers following basic visual features: color, texture, shape and etc. Then an analysis is made of the relation between CBIR and the MPEG-7 standard in this section.Section III on the basis of the studies of the MPEG-7 descriptors extraction method, makes an description of different weights method for different image regions with the 36-dimensionl quantification HSV color space; while in terms of the texture descriptor, the improved edge histogram descriptor is used.Section IV implements the image retrieved prototype system using two image retrieval methods based on color and texture discriptor. This section also gives the system design, system implement and the results of the experiment. In the end, this section analyses the results, which shows that two synthesized image feature retrieval is better than single one.