Adaptive Kernel Transform for Face Recognition Under Uneven Illumination Conditions

Face recognition plays an important role in person identification from a human face image. Existing approaches can extract facial features, but their performance is insufficiently accurate in uneven illumination conditions, especially in low light intensity. Therefore, an adaptive kernel transform (AKT) is proposed for recognizing face images under uneven illumination conditions. The proposed method provides horizontal-vertical detail components excluding illumination components by means of lighting component decomposition. Then it extracts robust features based on adaptive maximum histogram thresholding with matrix multiplication. The obtained features are composed of the important key facial regions without background noise and are robust in uneven illumination. From the Extended Yale-B face database, our AKT method yields 96.14%, which is higher than existing methods. The AKT method can adaptively extract robust features, thereby improving the effectiveness of face recognition under uneven illumination conditions.