A Combined Approach to SBIR using Edge Histogram Descriptor with Contourlet Transform

Abstract

This paper focuses on the problem of efficient and fast retrieval of images from a large database using sketch as query image. Basically searching is based on a descriptor that addresses the asymmetry between binary sketch from the user side and full color image of the database. The working of proposed algorithm is such that query image and full color database images undergo same feature extraction process. Database images will be clustered offline which reduces time complexity on runtime. Further indexing is done which will be used to describe, store and organize image information and to assist people in finding image resources conveniently and quickly. Firstly feature vector extraction is done using contours and then edges will be detected in different orientation using modulus maxima edge detection in contourlet domain. This approach is almost better than existing approaches in many aspects such as compactness of feature vector, simplicity of implementation, retrieval performance and efficient feature extraction less time complexity.