No abstract

Video face recognition is widely used for security surveillance and other applications in which the information about faces is extracted and processed. One of the problems usually present in video face recognition is to determine in real time the suitable images for the good performance of the algorithms, taking into account that although computers keep getting faster, the amount of information to process is higher than the capacity of image processing algorithms available. In this work we propose a method that allows to obtain in real time the best image of each person present in the scene from a sequence of images, considering both image and face characteristics, and using FPGA technology to accelerate the image processing. With the proposed implementation of the method we managed to process 37 times more images per second, and 97% of the selected images proved to be adequate for face recognition.