Improved Context Dependent logo matching framework using FREAK method

Abstract

Now days to prevent malicious use of original companies logos or identity, the automated image processing based frameworks are presented. The process of logo detection and recognition hence becoming the vital task for various applications. In this project we are presenting automated framework for logo detection using the real world logos images and its test image. Basically the working is that input query image is taken and big database of logos with goal of recognizing the logo in query image if any. Previously efficient method presented which outperform the existing method in terms of FRR and FPR. During this paper we are contributing by using RANSAC in which Fast Retina Keypoint (FREAK) descriptor is extracted for further matching and recognition process rather than using existing SIFT technique. The recent method for logo recognition and detection process is based on methodology of CDS (Context Dependent Similarity) which directly local features spatial context. Basically this CDS method using the SIFT method for initial keypoints extraction and then further matching process along with detection is done. The goal of our proposed CDS with RANSAC is to improve the recognition accuracy and to minimize the error rate performance. The RANSAC method is using FREAK technique for keypoints extraction which is superior as compared to SIFT.