Image content modeling and retrieval using sparse representation

Automatic image retrieval similar to a query image is an important task in computer vision. To obtain this goal, every image in database needs to be modeled. A signature is learned for each image and stored. A sparsity based image content modeling and retrieval is proposed in this paper. Sparsity based data modeling has been successfully applied across various areas of image processing. A dictionary is learned for each image in database. For image retrieval the query image features are extracted and sent to database of image dictionaries. Feature vector is checked against union of subspaces expanded by a dictionary. Each dictionary in database returns a score. Dictionary with highest score gets a vote in its favor. Image corresponding to dictionary receiving highest number of votes is said to be containing most similar content.