VGHN: variations aware geometric moments and histogram features normalization for robust uncontrolled face recognition

Abstract

The face recognition under the uncontrolled conditions is a widely debated research topic since from last decade due to technology advancement and the emergence of face recognition applications. The uncontrolled conditions such as illumination and pose variations, light intensity variations, etc. lead the poor face recognition performances. The extraction of invariant features in the presence of illumination variations is a difficult task. The variations of light intensity in face images effective in case of large-scale features that are truncated in recent techniques to release clarification mandatory specialties. But the loss of salient features during the release method of small-scale specialties leads to poor face identification performance. In this paper, the robust face descriptor method designed to discuss the challenges of face identification following uncontrolled environments. The proposed framework of variations aware geometric moments and histogram features normalization (VGHN) designed to handle the variations in illumination, pose, and light intensity of face images. In pre-processing, the difference of Gaussian filtering method applied to smooth these variations of each face image. To bridge the semantic recess within the spatial learning and histogram description, we build the face descriptor using the kirsch compass masks to extract the edge directional patterns from the pre-processed image. From each directional pattern, we extracted the variations aware and meaningful features using Geometric moments. The histogram features then removed from the pre-prepared face image. The rich set of features representation of face image has performed by the fusion of geometric moments and histogram features. The artificial neural network and support vector machine used at the end for face recognition and classification purpose. The representation of the VGHN system estimated working various research face datasets. The outcomes show that VGHN was able to improve the robustness compared to existing methods.