An Adjustable Face Recognition System for Illumination Compensation Based on Differential Evolution

It is well known that face recognition (FR) systems cannot perform well under uncontrolled conditions, but there are no general and robust approaches with total immunity to all conditions. Hence, we present an adjustable FR framework with the aid of the Differential Evolution (DE) optimization algorithm. This approach implements several preprocessing and feature extraction techniques aiming to compensate the illumination variation. The main feature of the present work stands on the use of the DE which is responsible for choosing which strategies to use, as well as tunning the parameters involved. In this case study, we aim to address the illumination compensation problem applying on the well known Yale Extended B face dataset. According to the proposed FR framework, the DE can choose any combination of the following techniques and tune its necessary parameters achieving optimized values: the Gamma Intensity Correction (GIC), the Wavelet-based Illumination Normalization (WBIN), the Gaussian Blur, the Laplacian Edge Detection, the Discrete Wavelet Transform (DWT), the Discrete Cosine Transform (DCT), and the Local Binary Patterns (LBP). Our experimental analysis confirms that the proposed approach is suitable for FR using images under varying conditions. It is proved by the average recognition rate of 99.95% obtained using four different datasets.