Research Of Face Recognition Algorithm Based On Gabor Transform

With the characteristics of intuitive and non-contact, Face recognition technology has become the focus in the field of bio-metric features recognition. So a large number of mature theory and algorithms about Face recognition were proposed. At present, there are some shortcomings in the face recognition under non-ideal conditions, and it is also to be further studied for face recognition under non-ideal conditions.Gabor transformation was extremely similar to the receptive field models of the mammalian retinal nerve cells, which is prevalently used in feature extraction. However, There are some disadvantages in traditional Gabor transform, such as DC component, limited bandwidth, high feature dimension, time consuming and so on. Log-Gabor transform for face feature extraction used in feature extraction, which has the general properties of Gabor transform and characteristics of containing no DC components, unlimited bandwidth, fast computing speed, etc. The face recognition algorithm based on sparse representation classification(SRC) has higher recognition rate and better robustness compared with face recognition algorithm based on subspace in illumination change and occlusion. The face recognition algorithm based on collaborative representation classification(CRC) has comparable recognition rate to SRC and faster computation speed. For the SRC and CRC algorithms, such as Eigenfaces, Randomfaces, Fisherfaces and other holistic features are mostly adopted, but these holistic features are not very effective for face recognition in illumination change, expression change, pose change, occlusion etc. The research of this paper is based on Gabor transformation and CRC algorithm. The main research results of this paper are as follows:?Aiming at the problem of Gabor feature and collaborative representation(Gabor-CRC)which has low recognition rate in illumination change, expression change, occluded face, a face recognition algorithm based on Gabor feature by blocks and collaborative representation(BG-CRC) is proposed. Firstly, the image is divided into some blocks and the Gabor features are extracted for each block. Secondly, The collaborative representation is used to compute thecategory of each block. Finally, vote for categories of all blocks to get the final category of the image. Through the simulation experiment, the BG-CRC have a better recognition rate than Gabor-CRC in illumination change, expression change and occlusion.?After the face image divided into blocks, each block has a different contribution on the face recognition. Through researching the contribution of each sub-block, a method of using sub-block recognition rate as the weight is proposed. And through simulation experiments,verify the effectiveness of the weight calculation method.?Log Gabor features combined with CRC algorithm in this paper, a face recognition algorithm is proposed based Log-Gabor feature and collaborative representaion. Feature extraction before dividing image or feature extraction after dividing image: which is more effective for face recognition? The face recognition algorithm based on Log-Gabor feature block and collaborative representation,the face recognition algorithm based on Log-Gabor feature block and weighted collaborative representation and the face recognition algorithm based on Log-Gabor feature by blocks and collaborative representation are proposed. Through experimental simulation and verify that feature extraction before dividing image has better recognition rate and robustness.?This paper validated the proposed algorithm in the self-built face database and the experiment achieved the desired effect. It could be regarded that these proposed algorithms had a certain practical value.