Research On Improved Recognition And Tracking System Based On Infrared Face

Using infrared face image as biotic feature to perform face recognition has become one of the hot points of pattern recognition. Especially, the infrared image could avoid the side effect of illumination variant, camouflage etc. To demonstrate the theory of dynamic image detection, recognition, and tracking based on the infrared image will be the main task of this paper. First, we collecting the infrared face images from the given infrared equipment, studying the infrared image in details, especially the request of image processing speed and accuracy under the dynamic circumstance. Then, aimed at the unique feature of infrared image, we constructed the algorithm of face image detection, recognition, and tracking. Finally, the experiments results tell that the proposed algorithm is practical and feasible.Considering the recognition and tracking performance is heavily dependent on the speed and accuracy of the image feature points detection, we adopted simple and effective classifier during the face image detection. For the accuracy of face detection, we attenuated the effect of face expression in large, and selected relative stable features to perform detection and position-setting. Before performing the face recognition and tracking, we should first generalization the images by scale and gray, and then process the images by image enhancing etc.During the recognition period, combined the K-L transform and the 2DPCA method, we introduced the blocks parted 2DPCA method, by advanced the former algorithm, then we bring out the weighted average 2DPCA method. By this algorithm, we addressed the side effect of face expression during the recognition period successfully.In the tracking period, we first analysis the mean shift method deeply, then compare the merit and drawbacks of the density estimation with parameter and the density estimation without parameter. After some research and analysis we find it is the accuracy of kernel function bandwidth that affect the tracking effect. We using this paper's face detection method to perform face accuracy poisoning, and using the poisoning ellipse to find the accuracy of the probability density kernel function's bandwith. And then we present a infrared face tracking method, this method is the combination of the mean shift method and the Adaboost method. From this way, we improved the tracking speed and accuracy, and realized real time infrared face.Finally we realized the proposed algorithm through C++ programme under the MFC. And the result was test by HID infrared face image set, and the result is comfortable.