Capture of meaningful faces from motion images

摘要

There is an increasing need and demand for automatic, efficient and reliable key information extraction from the video in industrial engineer. The face is one kind of the key information and usually the image is not clear in low-cost real-time surveillance system. The definition of the image mostly depends on the energy of the high frequency. The Fast Fourier Transform (FFT) and the Discrete Wavelet Transform (DWT) are two excellent transform tools. This paper proposes a method using FFT and DWT to analyze the face definition in motion images. The proposed method consists of three steps. Firstly, we preprocess the image by Histogram equalization and morphology. Secondly, we processes face detection and tracking. Finally, we capture the clearest face. We compare FFT with DWT, and experimental results show that DWT base on Bior3.1 is the most efficient.