Application of Adaptive Contrast Stretching Algorithm in Improving Face Recognition Under Varying Illumination Conditions

Abstract

This work proposed a novel algorithm, adaptive contrast stretching algorithm (ACS), in improving face recognition under varying illumination conditions. The ACS algorithm, whose building blocks are tuned logarithm filter and anisotropic diffusion filter (ADF), was used to preprocess samples of face images obtained from the extended Yale face database B. The resulting preprocessed data was split into training and testing datasets. While the training dataset was used to train a deep convolutional neural network (DCNN), the testing dataset was subdivided into four subsets based on the azimuthal angle of illumination. In order to compare the recognition accuracy obtained from using the ACS algorithm, the face images in the training dataset were successively processed using discrete cosine transform, difference of Gaussian, weber faces, multi-scale retinex and single-scale retinex. The respective output images obtained from each technique were used to train the DCNN. The result obtained from each technique showed that the developed ACS algorithm significantly outperformed other algorithms used in this study with an accuracy of 95%. This value is 2.5% greater than the unimproved version of the ADF, which is currently one of the acclaimed techniques used by most computer vision researchers in the surveyed literature.