Difference of Gaussian Oriented Gradient Histogram for Face Sketch to Photo Matching

S. A. Suandi,S. Setumin
DOI: https://doi.org/10.1109/ACCESS.2018.2855208
IF: 3.9
IEEE Access
Abstract:Retrieving a corresponding photo from a forensic sketch remains a great challenge. One of the difficulties is that the retrieval rate reduces significantly on data sets with shape exaggeration and lighting variation. The goal of this paper is twofold. First, we attempt to reduce the factor of the shape exaggeration problem by introducing new fiducial points for geometric face alignment. Next, to minimize the illumination effects, we describe the image using the difference of Gaussian-oriented gradient histogram. The matching results using the simplest distance measure on two public databases indicate that our proposed face sketch to photo matching method achieves significantly better accuracy than the state-of-the-art methods. It is also proved that shape exaggeration influences can be reduced by employing outer points face alignment.
Computer Science
What problem does this paper attempt to address?