Wavelet based multiresolution histogram for fast image retrieval

An important task in most content-based image retrieval (CBIR) systems is similarity matching. Similarity matching requires feature vector distance computation for each candidate image in an image database. Conventional algorithms based on exhaustive search are highly time consuming and inefficient. With the rapid increase in database size, there is a growing need for a fast and efficient retrieval system. A multiresolution data-structure based approach provides a good solution to the problem, but there is still scope for improvement. We propose a wavelet based multiresolution data-structure algorithm for faster image searching. The proposed approach reduces computation by around 50% over the multiresolution data-structure algorithm. In the proposed approach, we reuse the information obtained at lower resolution levels for similarity matching at higher resolution levels. This algorithm also saves disk storage space by about 50% over the multiresolution data-structure approach. The proposed approach can be easily combined with existing algorithms for further performance enhancement. We use the proposed approach to match similarity between luminance histograms for image retrieval.