Image Retrieval By Color, Texture, And Spatial Information

A novel approach to image retrieval using color, texture and spatial information is proposed. The color information of an image is represented by the proposed color homogram, which takes into account both the occurrence of colors of pixels and the colors of their neighboring pixels. The proposed Fuzzy Color homogeneity, encoded by fuzzy sets, is incorporated in the color homogram computation. The texture information is described by the mean, variance and energy of wavelet decomposition coefficients in all subbands. The spatial information is characterized by the class parameters obtained automatically from a unique unsupervised segmentation algorithm in combination with wavelet decomposition. Multi-stage filtering is applied to query processing to reduce the search range to speed up the query. Color homogram filter, wavelet texture filter, and spatial filter are used in sequence to eliminate images that are dissimilar to a query image in color, texture, and spatial information from the search ranges respectively. The proposed texture distance measure used in the wavelet texture filter considers the relationship between the coefficient value ranges and the decomposition levels, thus improving the retrieval performance. The final query ranking is based on the total normalized distance in color, texture, and spaThis research was supported in part by NSF CDA-9711582. tial information of all images passing the three filters. The experimental results show the effectiveness of the proposed