基于视觉运动特性的视频时空显著性区域提取方法

Motion Characteristics Based Video Salient Region Extraction Method

为了更准确有效地提取人眼观察视频的显著性区域,提出一种基于视觉运动特性的视频时空显著性区域提取方法。该方法首先通过分析视频每帧的频域对数谱得到空域显著图,利用全局运动估计和块匹配得到时域显著图,再结合人眼观察视频时的视觉特性,根据对不同运动特性视频的主观感知,动态融合时空显著图。实验分析从主客观两个方面衡量。视觉观测和量化指标均表明,与其他经典方法相比,所提方法提取的显著性区域能够更准确地反映人眼的视觉注视区域。

The human eyes only observe the salient region of the video. Thus a motion characteristics based salient re- gion extraction method was proposed. Spatial saliency map is extracted by analyzing the log spectrum of each frame in the frequency domain. Temporal saliency map is obtained by global motion estimation and block matching. According to the human visual characteristics and the subjective perception of different motion characteristics, the region of saliency is fused dynamically by spatial and temporal saliency map. The experiment was analyzed from both subjective and objective indicators. Visual observation and quantitative indicators show that the proposed method can reflect the human visual attention area more accurately than other classical extraction methods.