Stationary Wavelet-Based Adaptive Face Transform for Face Recognition in Lighting Conditions

Face recognition plays an important role in identifying an individual person. However, lighting conditions, especially low light intensity, are a serious problem which affects the recognition accuracy. The existing method is insufficiently accurate. Therefore, a stationary wavelet-based adaptive face transform (SWAFT) is proposed to extract features which are robust to lighting conditions for face recognition. Due to a property of the shift and translation invariances, stationary wavelet transform is applied to provide reconstructed images from horizontal, vertical, and diagonal detail coefficients which exclude lighting components. Then adaptive maximum histogram thresholding with matrix multiplication adaptively extracts robust features from reconstructed images. Two-dimensional principal component analysis is employed to reduce robust features by retaining the main information of features. Lastly, a support vector machine efficiently constructs the model and then classifies the robust features to identify each person. From the Extended Yale B database, the proposed SWAFT method achieves an accuracy of 97.81%, which is higher than conventional methods. Our method can adaptively extract robust face features, thus improving the effectiveness of face recognition in uneven lighting conditions.