Motion Saliency Detection based on Spatiotemporal Amplitude Spectrum Modulation

摘要

This paper focuses on motion saliency detection for video frames. At first, we define motion saliency as the uniqueness of spatiotemporal movement pattern, and reveal the underlying relationship between the three-dimensional amplitude spectrum and the moving object's velocity, direction and motion saliency. A modulation algorithm based on visual search asymmetry is proposed and applied in a multi-channel competitive structure to realize the quantitative measurement of motion saliency. Simulation results show that our model has better performance in detecting motion saliency, as compared with other competing models.