An Efficient Multi-filter Retrieval Framework For Large Image Databases

An efficient multi-filter retrieval framework for image retrieval in large image databases is proposed. Multiple filters are used to reduce the search ranges at different stages and thus save the time spent on unnecessary similarity comparison. First, a color label histogram filter uses a color label histogram with only thirteen bins to eliminate those images in the image database that are dissimilar to a query image in colors. Next, a wavelet texture filter discards the images that are dissimilar to the query image in texture from the query results of the color filter. A texture distance measure that considers the relationship between the coefficient value ranges and the decomposition levels is proposed. Finally, a spatial segmentation filter removes images dissimilar to the query image in spatial information from the query results of the texture filter. A unique unsupervised segmentation algorithm together with the wavelet technique produces the spatial features of an image automatically. All images passing the three filters are ranked based on the total normalized distance in color, texture, and spatial information. The top N images are displayed in the user interface. The experimental results demonstrate that the proposed framework dramatically reduces the search range.