Image Query by Multiresolution Spectral Histograms

We propose a method for searching a database for similar images guided by the image submitted by users as a query. We make use of texture features of images for the retrieval. To extract texture features, we apply complex multiresolution analysis to a query and database images and generate multiresolution spectral histograms from the coefficients of each subband image. We define an image query metric by inquiring the degree of analogy among the spectral histograms of a query and database images. The proposed method shows the improvement of the accuracy of query results against the conventional multiresolution method. Testing on the databases of 1800/2400 images, we obtain the good matches in 3.4/3.5 seconds