Researchs On The Techniques Of Semantic-Based Image Content Retrieval

With the development of information technology and multimedia technology, more and more images appear in our daily life. Therefore, how to retrieve data effectively from mass data of images that satisfy the user is an urgent problem to solve.Semantics-based image retrieval becomes a hot research field of image retrieval because it could retrieves images according to people's subjective understanding and sense, get rid of the hard tagging process and subjective inaccuracy from the traditional text-based retrieval to a certain extent in image retrieving, reduce the heavy dependence on underlying details of image in content-based image retrieval efficiently. However, there are big differences between underlying character of images and the understanding of people, which include that the underlying characters could not describe the high-level semantics content of image directly, there is a semantic gap between the underlying visual characters and the abstract semantics the image contains, image semantics have much complexity and fuzziness, and there are difficulties in extracting expressing and applying. All those make the technology of semantics-based image content retrieval becomes a challenging research subject.The key technology of the semantics-based image content retrieval are studying and extracting the underlying characters of image, studying the structure model based on similarity and matching, establishing the semantic network, defining several semantic rules for semantics, taking feedback training, studying and reasoning in human-computer interaction, to make better combination of underlying characters and semantic information, narrow and eliminate the semantic gap, implement the accurate image retrieval.This thesis firstly introduces extracting and matching of the image underlying characters and the semantic model, then studies extracting of the image semantics and description method, collects semantic information by using the feedback from human-computer interaction, establishes and trains the semantic network, improves retrieval technology and efficiency. And it designs and implements a prototype of semantics-based image retrieval system to the semantic classification and retrieval of image.At last, this thesis makes a summation to generalize the work, and looks forward to the difficulties, hot points and the problems that need to be resolved.