Image retrieval based on multi-scale edge model

Abstract

We propose a novel scheme for image retrieval using a wavelet based multi-scale edge model. All images in the database are decomposed into their multi-scale primal sketch and the background images respectively. The images are stored in the form of the extracted edge structures and background. The similarities between query image and the images in the database are measured based on the statistics of edges structures. The multi-scale edge modeling of image database can also be performed real-time to enable the image retrieval on arbitrary image databases. Experiment shows that the proposed scheme gives promising retrieval performance over the conventional retrieval methods.