ARM9-Based Face Detection System Using Power Spectrum of Fourier Transform

In this paper a novel hardware implementation of a face detection system of frontal faces with various backgrounds is proposed. In the first stage of the face detection algorithm, 40×40 windows are extracted from image and high frequency components are removed by two dimensional Fourier transform, then the image will be transformed to time domain by two dimensional logarithm power spectrum of Fourier transform .For dimensionality reduction and decreasing the CPU processing load in classification stage the traditional PCA algorithm is used. Finally the Input parameters from PCA algorithm are applied to a Multi Layer Perceptron (MLP) neural network which is trained using boot-strap algorithm to classify the face candidates and non-faces. Two type of test has been done to evaluate the efficiency of proposed algorithm and speed of system. At first the algorithm evaluated on a PC system on CMU and BioID test sets. Then the algorithm implemented on designed hardware with an ARM9 processors. Experimental result shows that the algorithm has better throughput in detecting faces while reducing the false alarms. Also using ARM9 processor makes system faster and can be used as a real-time face detector system in small-size images

کلیدواژه ها:

component ، Face Detection ، Fourier Transform ، MLP ، Boot-Strap ، PCA ، ARM9