Binary Gradient Correlation Patterns for Robust Face Recognition

This paper presents a computationally efficient yet powerful binary framework for robust facial representation based on image gradients. It is termed as binary gradient correlation patterns (BGCP). To discover underlying local structures in the gradient domain, BGCP computes image gradients from multiple directions and simplifies them into a set of binary strings. Certain types of these binary strings, defined as ”structural” patterns, resemble fundamental textural information of images. They detect (micro) orientational edges and provide strong orientation and locality capabilities, thus enabling great discrimination. BGCP also benefits from advantages of the gradient domain and exhibits strong robustness against gray-scale variations, e.g. illuminations. The binary strategy realized by pixel correlations in a small neighborhood subsequently simplifies the computational complexity, with only 0.0032s processing time for a typical face image. Furthermore, the discrimination power of BGCP can be enhanced by applying it to a set of defined orientational image gradient magnitudes, further enforcing locality and orientation. Results of extensive experiments on various benchmark databases illustrate significant improvements of BGCP based representations over thestate-of-the-art methods in the terms of discrimination, robustness and complexity. Codes for BGCP methods will be available at http://www.eee.manchester.ac.uk/research/groups/sisp/software/.