An Optimized Illumination Normalization Method for Face Recognition

Abstract

Differences in illumination conditions cause significant challenges for any 2-D face recognition algorithm. One of the methods to counter these effects is image preprocessing before feature extraction. In this paper we present a new preprocessing approach that uses custom filters obtained through an optimization procedure striving for most suitable preprocessing filters for the selected feature extractor and distance measure. We experiment with it using Local Binary Pattern texture features and x2 histogram distance metric. Results are provided for Face Recognition Grand Challenge (FRGC) 1.0.4 dataset.

Augmented SRC for face recognition under quality distortions

“…In (16), \ and mod correspond to the integer division and modulo operations, respectively, and l = 0 corresponds to the absence of distortion. We apply the SRC using the proposed dictionary to classify test samples based on the class-wise minimum reconstruction error as in (5). Nevertheless, as the subject belongs to either one of the N d possible levels (including the absence) of distortion, the subject identity i is obtained as follows:…”

“…This may greatly affect the performance of face recognition algorithms. While face recognition has already achieved a very good performance over large-scale galleries that include traditional scene-dependent distortions such as pose variations, extreme ambient illumination [2][3][4][5][6][7], and partial occlusions due to obstacles and disguise [7][8][9][10][11][12], there still exist many more challenges related to variations in image quality such as blur and block occlusions due to packet loss. These types of distortions, which result from capture, processing, and transmission are commonly present in images and videos acquired by surveillance cameras and increasingly by mobile handheld devices or transmitted over Internet protocol and wireless networks.…”

Illumination Processing in Face Recognition

“…Holappa et al (Holappa et al, 2008) presented an illumination processing chain and optimization method for setting its parameters so that the processing chain explicitly tailors for the specific feature extractor. This is done by stochastic optimization of the processing parameters using a simple probability value derived from intra-and inter-class differences of the extracted features as the cost function.…”

Face Identification Using Large Feature Sets

“…Among the best known illumination normalization methods are the self-quotient image (SQI) [46], total variation models, and anisotropic smoothing [25]. SQI is a retinex-based method, which does not require training images and has relatively low computational complexity; we use it due to its simplicity.…”

“…They used local ternary patterns and a Hausdorff-like distance measure. Holappa et al [25] used LBP texture features and proposed a filter optimization procedure for illumination normalization. Aggarwal et al [26] presented a physical model using Lambert's law to generalize across varying situations.…”