Multi-mode Image Retrieval Method

Content-based image retrieval is an image retrieval technology based on the visual features (color, texture, shape, etc.). With the computer science and the database technology development, CBIR has become a hot spot at home and abroad, has been widely used in the field of meteorology, remote sensing and medical imaging.In this thesis, the background, purpose and meaning of content-based image retrieval are introduced. We review briefly the research status and development of content-based image retrieval technology at home and abroad. Some relative key techniques of image retrieval are list out; also the system's frame construction is given. Color has been widely used as important visual image information. In the research of images'color features, this thesis is devoted to the description and mutual conversion of color space models, quantification of color, and the different methods of color feature extraction. The similarity measurement is introduced, and some means of features matching are put forward. Their performance is analyzed and compared by results of the simulation.As for the shape feature, the image enhancement is deeply studied, and the traditional edge extraction methods are reviewed, also the different kinds of description and extraction means of images'shape features are discussed. An improved OTSU algorithm based on the Rayleigh model, as well as a combinational features extraction using shape and structure characters, is proposed in this thesis.Finally, based on the above-mentioned theory, through combining the two features of color and shape, a multi-feature and multi-mode image retrieval system is designed, and implemented by using Visual Studio.Net programming. The validity of system is verified by simulation. The results show that, through comparison of data obtained in image indexing results, regardless of the precision and recall rate, the multi-mode image retrieval system's results is significantly superior to the ones of other systems employing single kind of feature.