Frequency spectrum modification: A new model for visual saliency detection

Abstract

Previous research has shown that Fast-Fourier-Transform based method was an effective approach for studying computational attention model. In this paper, a quantitative analysis was carried out to explore the intrinsic mechanism of FFT-based approach. Based on it, a unified framework was proposed to summarize all existing FFT-based attention models. A new saliency detective model called Frequency Spectrum Modification (FSM) was also derived from this framework. Multiple feature channels and lateral competition were applied in this model for simulating human visual system. The comparison between FSM and other FFT-based models was implemented by comparing their responses with the real human eye's fixation traces. The result leads to the conclusion that FSM is more effective in saliency detection. © 2010 Springer-Verlag.