Research Of Face Image Enhancement Method Based On Sequence Image Fusion

At present, the electronic video surveillance system is widely used intraffic monitoring and scheduling, security monitoring inside and outside of thebuilding, security control of the banking system and so on. In the application ofsecurity monitoring, the face recognition based on sequence image is oftendynamic images recognition, and its characteristic is that pose, glasses,accessories, illumination, imaging angle and the imaging distance is random anddynamic, which is affected the accuracy of face recognition. In order to solvethis problem, seriously considering the interference of illumination and poses,this paper proposed a face image sequence enhancement algorithm based ondynamic image sequence information form face detection, normalization, imagepretreatment and image fusion. First of all, the face detection method based on skin color model andgeometric feature was used to make the face separated from the sequence imageand get rid of the background, hair, clothes and other redundant information.Secondly, different methods were used to realize gray normalization accordingto the face image difference in the gray level resulted form different lightconditions; and geometric normalization based on eyes location was completedaccording to the difference in the geometric location resulted from differentposes and scale; then in order to evaluate the results after normalization, thecorrelation coefficient was adopted to make similarity detection on the sequenceimage after normalization; And then, the sequence of face image afternormalization was fused by three different fusion rules based on wavelettransform. Finally, image quality evaluation for the result of image enhancementwas made by the subjective and objective evaluation method.When it came to validating the method of this enhancement based onsequence image fusion, three kinds of face image and three fusion rules wereused. The three kinds of face image were got form in-plane rotation, in-depth rotation and different light conditions. The three fusion rules were the rule basedon largest absolute value of coefficient, the rule based on largest absolute valueof sum in neighborhood, and the rule based on largest value of variance inneighborhood. The results show that: the image enhancement method based onsequence image fusion has realized face detection in the condition of differentbackground, hair, poses, scale and completed the normalization of face imagesequence, also it has taken full advantage of the information in more images tomake up the lake of single image and realize image enhancement; when thenumber of original images is small and the light condition is normal, the rulebased on the largest absolute value of coefficient to the image enhancement ismore effective and more rapid; when the number of original images is more, andeven if both the poses and light condition changes, the rule based on largestabsolute value of sum in neighborhood to the image enhancement shows someadvantage in the enhancement and running speed.