An optical information processing-based idea for visual attention analysis

One of the challenges in visual attention analysis is to generate full-resolution saliency map in real time. Traditional methods usually design algorithms and implement in computer, and they have faced a troublesome problem in balancing the resolution and computational speed. Optical computing is considered to be a practical and efficient method, so we build an optical information processing-based system to achieve fast visual saliency extraction to generate full resolution saliency maps with fine details of boundaries. The whole system mainly consists of a Display as the input image source, optical Fourier Lens and spatial filters, an imaging sensor. The key of our method is frequency selection with two precise pinholes which treated as band-pass spatial filters. And the saliency map is finally obtained using simple calculation. According to the experimental results, and comparing with other famous approach, the results demonstrate that our system efficiently produce full resolution saliency maps with good boundaries.