Research On Face Alignment Algorithm

With the development of the technology of the computer, some biological characteristicsof the human were applied to the authentication. Relative to other technology of biologicalrecognition, face identification has the advantages that feature collecting is more convenient,recognition cycle is shorter, etc. Face recognition system can be divided into the followingseveral parts: Face detection, face alignment, extraction of face feature and face classificationand recognition. Alignment of different faces’ image in face sequence image and positioningand extracting of the face feature become the most important question of the face recognitionsystem. In order to provide a strong robusticity method of the face alignment, we did theresearch.Generally speaking, the realization of the face alignment can be divided into tworelatively independent process: Face detection which could judge a images, to be detected,whether comprise the face or not; Face alignment which could fixed the position of the faceprecisely based on the face detection. Face detection was used in this research, which wasbased on Haar-like feature. This method has the advantage that good effect, high accuracytesting, high speed of detection, etc. The face image, which were obtained by the method,were the preparatory work of the alignment algorithm.In this research,two kinds of methods of the face alignment were put forward: one isface alignment, which was based on the matching on SIFT feature algorithm;the other is themethod of template matching alignment,which is based on gradient image. Alignmentalgorithm for template matching, which was based on the gradient image, detected the imageby face detection, and calculate gradient image. This alignment method made face alignmenteffect achieved better. In the experimental process, this algorithm could make the execution ofaligned operation more quickly, and can generate aligned pictures in different size, accordingto the kind of the templates of face image.SIFT feature algorithm bring about that given a set of unaligned examplars of a faces, we automatically build an alignment mechanism, without any additional labeling of parts orposes in the data set. Using this alignment mechanism, new members of the faces resultingfrom a face detector, can be precisely aligned for the recognition process. This alignmentmethod improves performance on a face recognition task, both over unaligned images andover images aligned with a face alignment algorithm specifically developed for and trained onhand-labeled face images.In this research, according to the two kinds of alignment algorithm, we can get the facealignment image through the experiment. Meanwhile, comparing with the standard faceimage, getting error values and identification accuracy, we can obtain the face alignmentalgorithm with a stronger robusticity.