New Image Retrieval System Based on Robust Color and Texture Pattern Feature Extractions

In this dissertation, a new image retrieval system based on robust color and texture pattern feature extractions is investigated. The substance of research is composed of three main parts. First, robust feature extraction method from an original image based on advanced image processing techniques is proposed. The robustness designates geometric distortions (rotation, scaling and translation), illumination degradation and additive noise. Next, a construction method of efficient feature space suitable for image retrieval and classification is proposed. The feature space is obtained by a linear transform of the original robust feature space. The linear transform is generated by nonlinear least square optimization for the membership matrix. Further, by unifying the color and texture features, the user friendly fuzzy distance that fits the sense of distance is obtained. In the computer simulations, the individual process is examined. The effectiveness of the methods is verified in comparison with the conventional approaches