A Method For Spatially Weighted Image Brightness Normalization For Face Verification

摘要

Despite the major advances, the accuracy of modern face verification systems depends on the lighting conditions. Thevariability of illumination can be compensated either by performing image preprocessing or by training more robust verificationmodels. Nowadays, great priority is given to the development of neural network classifiers, while the importanceof image preprocessing is being undeservedly neglected. This article proposes a method for spatially weighted brightnessnormalization of grayscale face images which preserves the relevant image information. An experimental study is performedto demonstrate the effects of various methods for brightness normalization on the accuracy of the neural networkclassifier in the application of face verification. It is shown that brightness normalization can improve the face verificationaccuracy for images captured in complex illumination conditions, that is, to compensate for samples that were not fullypresent in the training data.