Recently, deep learning techniques to learn effective feature representations have swept a variety of computer vision tasks including face recognition with illumination, poses, and expressions problems. Thus, they can get some superior performance when the optimal thresholds are selected, compared with other noise-resistant methods. In the NRLBPs (NRLBP, NRLBP+, NRLBP++), more information of other bits and the prior knowledge of images are incorporated into the encoding process. In, the authors propose a mechanism to recover the corrupted image patterns in the original LBP. Noise-resistant LBP (NRLBP) and its improved versions (NRLBP+, NRLBP++) are another kind of method to solve the noise-sensitive problem.
However, given the magnitude of the pixel difference used in the calculating process, the FLBP algorithm is still sensitive to noise. For example, fuzzy local binary pattern (FLBP) is proposed to reduce the influence of noise which utilizes the probability measure to encode the pixel difference as 0 or 1. To address this issue, many methods are presented for direct recognition of the identity from the noisy image. But the denoised image tends to lose some of its edge information which hurts the image recognition in the subsequent stage. Both of the two categories of approaches can produce some good quality images. Another thread of methods is to capture image statistics directly in the image domain. A line of approaches is to transfer image signals to an alternative domain where they can be more easily separated from the noise. Various methods have been proposed to denoise the image before the recognition stage. And sometimes, it is even difficult to recognize an identity from the seriously noisy face by human. Face image is vulnerable to noises during its acquisition, quantization, compression, and transition. But when it comes to the noisy images, the recognition accuracy of most approaches would drop significantly. Extensive works have been carried out towards the illumination, pose, and expression problems and also get some excellent results.
However, in some uncontrolled conditions, including varying illumination, poses, facial expressions, and noise, the performance of face recognition system would be dramatically affected. Nowadays, face recognition has made great progress for various potential applications in security and emergency, law enforcement and video surveillance, access control, etc.
0 kommentar(er)
