The Application Of The Extended Local Binary Mode Algorithm Based On Gabor Filter In Video Face Recognition System

Biometric identification technology has gradually entered the line of sight of people, with the rapid development of computer vision, in recent years. Face recognition as a hotspot in biometric identification technology has been more and more attention, because of its unique advantages. However due to the objective cause such as the structure of face image's complexity, the sensitivity to external factor and so on, the research of face recognition technology is doomed to be a difficult and challenging topic. Local binary pattern(LBP) was proposed more than 20 years ago, and because of it's characteristics that it could describe Local texture feature simply and effectually, LBP has been as a research hotspot in the field of computer vision. The local binary pattern were applied in the field of face recognition have also made good effect, but local binary pattern operator structure is still not perfect. For example:the operator neighborhood is not universal, and the described information are imperfect. So, the researchers gradually put forward a series of improved LBP algorithm, including ILBP, LTP, LLBP and so on. This paper is to use an improved algorithm, Elongated local binary pattern (ELBP) for texture description of gray level range.There is a method based on Gabor wavelet among all kinds of algorithms. After the researches, researchers found that the human each optic nerve cells is equivalent to the Gabor filter with a certain scale and direction, And the process of human perception of visual images is similar to the convolution of images and Gabor filter. So that the method based on Gabor filter can effectively extract the multi-scale and multi-direction characteristics of face image.There are research work of this paper as following:1. A new algorithm that coalesced multi-scale and multi-direction Gabor amplitude characteristics with texture features of Elongated local binary for feature extraction pattern was proposed. After making the input human face image through Gabor filters, the ELBP features of Gabor amplitude images were extracted to form the histogram sequence, then cascading the histogram sequence with average maximum distance gradient amplitude to get the final features information. As mentioned above, this paper realized the effective description for multi-scale and multi-direction anisotropic information.2. Doing an test experiment at YALE, YALE-B, the UCD-VALID, CMU-PIE face databases and so on, and comparing with other mainstream algorithms. Finally proving that the proposed method can be more excellent for face recognition.3. This paper projected a real-time video face recognition system based on Visual Studio 2013 and OpenCV 2.0. The system do the face detection for the captured video, and then made the detected face finish operation like entering, identification and so on. The implementation of the system provides technical reserves and practical experience for me to apply face recognition technology to practical application in the future.