Sketch Based Image Retrieval in Large Databases using Edge Features

Abstract

Sketch-based image retrieval (SBIR) presents better flexibility in expressing the query as sketch for retrieval of images as opposed to text based retrieval. Using a sketch, it is easier to express the orientation and pose of the objects for image retrieval from the database. We propose an efficient approach for SBIR from large databases based on hand awn rough sketch. In the proposed method, images are synthesized to yield a binary sketch that is processed in similar way to user drawn sketch. Edge features are extracted by overlaying the sketch with non-overlapping and overlapping grids, respectively. The most similar images to the query are then retrieved from the database using weighted based similarity approach. Experiments are performed on flickr15k dataset yielding excellent retrieval performance in comparison to the methods available in the literature.