Recognizing surgically altered faces using local edge gradient Gabor magnitude pattern

For humans, every face is unique and can be recognized amongst similar faces. This is yet to be so for machines. Our assumption is that beneath the uncertain primitive visual features of face images are intrinsic structural patterns that uniquely distinguish a sample face from those of other faces. In order to unlock the intrinsic structural patterns, this paper presents in a typical face recognition framework a new descriptor, namely the local edge gradient Gabor magnitude (LEGGM) descriptor. LEGGM first of all uncovers the primitive inherent structural pattern (PISP) locked in every pixel through determining the pixel gradient in relation to its neighbors. Then, the resulting output is embedded in the pixel original (grey-level) pattern using additive function. This forms a pixel's complete structural pattern, which is further encoded using Gabor wavelets to encode the frequency characteristics of the resulting pattern. From these steps emerges an efficient descriptor for describing every pixel point in a face image. The proposed descriptor-based face recognition method shows impressive results over contemporary descriptors on the Plastic surgery database despite using a base classifier and without employing subspace learning. The ability of the descriptor to be adapted to real-world face recognition scenario is demonstrated by running experiments with a heterogeneous database.