Face recognition algorithm on the basis of truncated walsh-hadamard transform and synthetic discriminant functions

Abstract

This paper addresses face recognition algorithm by means of synthetic discriminant functions in reduced dimension space on appearance-based approach. Being apart from ideal conditions, where standard illumination, expression and frontal view are the normal case, we intend to process available information on the classic manner, providing useful and promising results. Half-tone facial image is 3D intensity shape, which we represent as being composed of a set of 2D binary images. Normalized gray image is sliced on the layers which are further converted to binary maps. 2D truncated Walsh-Hadamard Transform is then applied to these maps leading to significant reduction of dimensionality and producing translation-invariant feature vectors. These vectors are used to construct synthetic discriminant function (SDF), which is considered to be the facial image class descriptor and serves for face identification.

An efficient illumination normalization method in a transformed domain

“…The main difference between our method and the methods in [6][7][8][9] is that the low-frequency parts of WHT coefficients are discarded to eliminate illumination variations in our method. It is the first time that the WHT is applied in illumination invariant face recognition.…”

“…Because of its advantage of lowcomplexity, it has also been used in many signal processing and data compression algorithms, such as HD Photo and MPEG-4AVC. However, in face recognition, it has limited applications [6][7][8][9]. In [6], a face image was decomposed into a set of binary images, and 2D truncated WHT was applied to the binary images to reduce data dimensionality and to produce translation-invariant features.…”

“…However, in face recognition, it has limited applications [6][7][8][9]. In [6], a face image was decomposed into a set of binary images, and 2D truncated WHT was applied to the binary images to reduce data dimensionality and to produce translation-invariant features. Guo et al [7] proposed a novel face feature representation based on multi-scale 2D WHT in face detection.…”