基于运动和空间方向一致性的视频显著性检测

VIDEO SALIENCY DETECTION BASED ON CONTRASTING COHERENCE OF MOTION AND SPATIAL DIRECTION

根据人眼的视觉特点,提出一种基于对比运动和空间方向一致性的算法,用来检测视频序列的显著性区域。在该算法中,首先针对每一帧图像利用sobel算子计算其水平梯度、垂直梯度,然后用光流法计算运动矢量;接着构建三维张量结构,得到每个像素点与周围点的水平、垂直梯度和运动矢量一致性,并对比每个像素点与周边点的一致性;最后把对比结果转化为视频显著图。将该方法与已经存在的方法进行比较,实验结果表明,该方法可以很好地排除背景纹理区域的运动所带来的影响,能够准确地检测出各种运动场景的显著性区域。

According to the characteristics of human visual system( HVS),we proposed an algorithm to detect the salient region of video sequence,which is based on contrasting the coherence of motion and spatial direction. In this algorithm,it first uses Sobel operator to calculate the horizontal gradient and vertical gradient for each frame,and then calculates the motion vector by the method of optical flow.After that,it builds a 3D tensor structure to obtain the coherence of each pixel with its surrounding pixels in regard to the horizontal and vertical gradients and the motion vector,as well as contrasts the coherence of each pixel with its surrounding pixels. Finally it converts the contrasting results to video saliency map. Comparing this algorithm with existing methods,experimental result showed that it could well exclude the impact of motion in background texture region and accurately detect the salient regions of all kinds of motion scenes.