Face Recognition Based On Local Binary Pattern And Transform Domain

Face recognition involved in pattern recognition, machine vision, imageprocessing and many other disciplines, has important theoretical research value andbroad application background. Face image exists in a variety of conversion factors ofthemselves and the environment, so face recognition is complex and difficult, andmany key technical problems remain to be further addressed and improved. This papermainly in-depth study and research the illumination preprocessing, Local binarypattern, wavelet transform, Curvelet transform and image processing technology, inthis paper facial feature extraction is carried out for a detailed analysis and a newalgorithm, the main work includes the following aspects:Propose a feature fusion approach based on wavelet transform and multi-scaleLBP for face recognition. Firstly, do illumination preprocessing to face images andthen block them, on different sub-blocks, different scales of LBP operator were usedto extract the histogram feature vectors, link up LBP histogram features of eachsub-block as the LBP feature vectors of this face image; next do waveletdecomposition to face images, extract wavelet feature vectors, and finally the LBPfeatures and wavelet features with weighting are fused to produce the feature spacefor classification. Experimental results show that the recognition rate after featurefusion compared to the recognition performance of single feature has someimprovement.Proposes a face recognition method combined Curvelet Transform and LBPOperator for the low frequency coefficients that the illumination affects Curvelettransform, and the high-frequency coefficients in facial expression transformation andCurvelet transform has a directional sensitivity, firstly, do the illuminationcompensation to face images, and then Curvelet transform the preprocessed faceimages, by modifying the optimal scaling factor layer and layer of fine-scale toachieve the purpose of denoising the image and enhancing the texture, using theinverse transform to reconstruct face images, multiple layers of the reconstructedimage block, on the different sub-block use multi-scale LBP operator extract featurevectors for face recognition, compared with that face recognition which combinestraditional wavelet analysis and LBP operator, Experimental results show that it hasachieved a better recognition effect in the case of changes in illumination and expression.