Researches On Object Location And Face Recognition Based On Symmetry Information

Object localization is the premier step of some complicated systems such asobject tracking and recognition, the accurate processing and real-time developing ofwhich have become one of the key techniques in modern image processing field.There exist numerous symmetric objects in real world. These are the research subjectsof many complicated systems such as the car in intelligent transportation systems andthe face in face recognition systems. Therefore, symmetric object localizationtechnology has very high academic research value and broad application prospects.This paper proposes an effective object location method based on object symmetryproperty. Face recognition is a kind of typical object recognition and a hotspot in theresearch of computer vision and pattern recognition field. However, the pose variationproblem is still one of the difficulties in current face recognition. Face is one of themost common symmetric objects. For two-dimension face reconition problem undermultiple poses, this paper proposes a new face recognition method based on facesymmetry information which is nonsensitive to pose variations. Main researchachievements are as follows:1. Object localization algorithm based on symmetry information: This paperproposes a new feature descriptor named the symmetric SIFT descriptor based onobject symmetry characteristic research. It is improved from SIFT method against thedefect of which can only detct symmetry axis of object patch with rich texture insymmetry detection application. The proposed descriptor can be used to describe theedge feature point, broking through the limit of SIFT method in the aspect ofdescribing edge feature points. This paper proposes a object detection method basedon symmetric SIFT descriptor to solve symmetry object location problem. In thismethod, the object’s complete symmetry axis is first extracted and the object locationis then accurately finished.2. Face recognition algorithm based on symmetr y information: Centering aboutthe problem of face recognition of multi-pose in two dimentions, three facecorrelation degree evaluation indexs are put forward based on the prior knowledge ofsymmetrical nature of huma n face. Use these indexs to evaluate the correlation degreebetween the test sample face and training sample face to filter out training imageswith low correlation degree, the process of which is called coarse classification. Then, ASIFT method is employed to match the test sample face with the rest training samplefaces to obtain the final recognition results. The experiments results show that theintroduce of face correlation degree evaluation index based on symmetr y informationcan not only greatly reduce the burden of subsequent match, but also be helpful inimproving the overall recognition rate.3. Using the Microsoft vc++6.0to realize the software of object localizationalgorithm based on symmetry detection and the software of face recognition algorithmbased on symmetry.Finally, innovations and deficiencies of the work of this paper were summarized,and the work needed to be carried out in future research was pointed out.