Enhanced human face image searching system using relevance feedback

This paper addresses the problems of face recognition using sketch. All existing face sketch recognition systems focus on the sketch to mug shot matching. However, one of the key problems is that very often witness cannot reconstruct the sketch well. In turn, it does not look like the mug shot image. The performance of existing systems are greatly degraded. To overcome this limitation, this paper makes use of the concept of human-in-the-loop and proposes a human face image searching system using relevance feedback. The proposed system employs linear discriminant analysis for on-line learning the optimal projection subspace for face representation. The proposed system has been evaluated using FERET database and a Japanese database with hundreds of individual with all frontal view face images. The results are encouraging.