A Fast Computation Algorithm of Binocular Energy Model

This paper presents a fast computation algorithm of binocular energy model (BEM) to estimate the global disparity between a pair of stereoscopic images. The previous methods estimated the global disparity by using the convolution of receptive field function and stereoscopic images. But, it is known that the above approaches take a long time to obtain the required output. Hence, in this paper, we propose a fast computation algorithm by replacing the convolution with a multiplication operation in frequency domain. We can transform the image and the receptive field function into the frequency domain with the well-known fast Fourier transform called FFT. Then the output of the linear neurons is calculated simply with the inverse FFT (IFFT). The remaining process is the same as the previous approaches. With experimental results, we show that the proposed method can obtain the same global disparity as the previous schemes while reducing the computing time to 1/5 approximately.