CONTENT-BASED IMAGE RETRIEVAL USING ENHANCED HYBRID METHODS WITH COLOR AND TEXTURE FEATURES

Content-based image retrieval (CBIR) automatically retrieves similar images to the query image by using the visual contents (features) of the image like color, texture and shape. Effective CBIR is based on efficient feature extraction for indexing and on effective query image matching with the indexed images for retrieval. However the main issue in CBIR is that how to extract the features efficiently because the efficient features describe well the image and they are used efficiently in matching of the images to get robust retrieval. This issue is the main inspiration for this thesis to develop a hybrid CBIR with high performance in the spatial and frequency domains. We propose various approaches, in which different techniques are fused to extract the statistical color and texture features efficiently in both domains. In spatial domain, the statistical color histogram features are computed using the pixel distribution of the Laplacian filtered sharpened images based on the different quantization schemes. However color histogram does not provide the spatial information. The solution is by using the histogram refinement method in which the statistical features of the regions in histogram bins of the filtered image are extracted but it leads to high computational cost, which is reduced by dividing the image into the sub-blocks of different sizes, to extract the color and texture features. To improve further the performance, color and texture features are combined using sub-blocks due to the less computational cost