Research On Local Facial Texture Feature Extraction Method And Its Application

LBP is a operator to measure and extract local texture feature information in the image neighborhood,which has been widely used in facial feature description and face recognition because of its simple calculation and insensitive illuminations.However,due to the influence of non uniform illumination,noise and partial occlusion and etc on the image in the scene of practical application,which can lead to the degradation of image quality,original LBP feature extraction method can not meet the needs of face tracking and recognition applications.In order to improve the accuracy and robustness of face tracking and recognition further,the paper based on the practical application of face tracking and recognition is devoted to using the theory of LBP operator to come up with a new facial local texture feature method which is more robust to non-uniform illumination,noise and partial occlusion and etc than original LBP.the paper has made the related work and research are as follows.(1)In order to solve the shortcoming that LBP is sensitive to non-uniform illumination,the paper proposes a local relative differential binary pattern with adaptive threshold.Based on the method of relative difference transform,LBP with adaptive threshold is used to extract the facial local texture feature.Experimental results show that the proposed method is robust to extract the face local texture feature on the condition of non-uniform illumination.(2)For the problems that LBP can not extract the global mode information of local texture feature and be sensitive to noise,By combining the simulated retinal sampling structure(Patch to Pixel,PTP)and the center symmetric local relative difference binary pattern,the paper proposes a new local texture feature operator called PTP center symmetric local relative differential binary pattern with adaptive threshold,which can not only capture the the global mode information of local texture feature,but also can capture the micro-mode information.Comparative experimental results show that the proposed the operator has the anti-noise performance,while maintaining the robustness of the illumination.(3)For the problems that traditional face tracking algorithm based on mean shift only depends on using RGB color histogram to extract facial feature which can lead to the loss of face target by the influence of illumination change,and the similarity between the target and the scene.A mean shift face tracking algorithm based on RGB color feature and PTP_CSLRDBPAT texture feature fusion is proposed,Experimental results indicates the proposed method has better accuracy and robustness of face tracking when the illumination changes are obvious and the face target is similar to the background color.(4)In order to solve the problem that it is easy to cause the recognition accuracy to decrease when the face image is partially occluded.The paper proposes a new face recognition method based on local texture feature of monogenic signal transform.Firstly monogenic signal transform is used to decompose the face image into three feature maps for monogenic amplitude,phase and orientation.Secondly,for the monogenic amplitude,PTP_CSLRDBPAT is used to extract the local texture feature map,then for the monogenic phase and orientation,Local XOR Pattern(LXP)is used to extract the local texture feature maps.Thirdly,each three local texture feature map is divided into several sub-regions,and each sub-regions of the maps are set as the spatial saliency which is allocated to the sub-regions as different weights.Then weighted feature histograms of all the sub-regions of each three local texture feature map are connected together,Finally the three local texture features are weighted to produce the final facial local texture feature and K-Nearest Neighbor classifier is used to achieve face recognition.Experimental results show that the proposed algorithm can effectively improve the face recognition rate when the face image is partially occluded.