Illumination invariant face recognition: A survey of passive methods

Abstract

Face recognition under varying illumination is one of the challenging problems in real-time applications. Numerous methods have been developed by the research community to handle the problem. Existing surveys of methods are either too old or do not cover performance analysis of illumination invariant methods. This paper is more extensive than previous surveys and covers recently developed methods. The paper focuses on passive methods which solve the illumination problem by investigating the visible light images in which the face appearance has been altered by varying illumination. The methods are classified into four broad categories, namely (1) subspace-based statistical methods (2) illumination invariant representation methods, (3) model based methods, (4) other illumination handling methods. The other illumination handling category includes the methods which do not fall under first three categories. Performance analysis and discussion of methods and an evaluation of results is presented to determine the suitability and applicability of the method(s) for specific applications. It is observed from the survey of methods that illumination invariant representation based methods are better in terms of the number of training images required, the simplicity, computational complexity and robustness.

Robust Face Recognition by Computing Distances From Multiple Histograms of Oriented Gradients

“…A known fact in a face recognition task is that the differences of the face images of the same person (intra-class) can be much bigger than differences of the face images of different subjects (inter-class) due to different poses and illumination conditions [2], [3]. For example, two photos of the same person taken in different poses or illumination conditions will have a higher geometrical distance than two photos of two different people whose pose and illumination conditions are the same.…”

Face recognition under varying illumination based on gradientface and local features

“…It has also been revealed by the Face Recognition Vendor Test (FRVT) 2006 [4] that illumination variation is among the several bottlenecks for a practical face recognition system. Many methods have been proposed to deal with the illumination problem, which could be classified into three main categories [5,6]: preprocessing and normalization techniques [7][8][9], face-modeling-based approaches [10][11][12][13], and invariant-featurebased approaches [14][15][16][17][18].…”

“…The third category of approaches attempts to extract illumination‐invariant or illumination‐insensitive features for face recognition. Invariant‐features‐based methods are more effective compared to the above two categories , and do not demand any learning process. Quotient image (QI) was proved dependent only on the relative surface texture information and designed for dealing with illumination variation.…”

Face recognition by decision fusion of two-dimensional linear discriminant analysis and local binary pattern

“…To tackle the problems induced by the impacted negative factors, many approaches for face recognition of 2D images have been proposed, as described by the surveys [5,6]. To handle varying illumination problems, the self quotient image was used in Ref.…”