基于颜色和运动空间分布的时空显著性区域检测算法

针对复杂背景和运动条件下视频显著性区域检测准确度不高的问题,本文提出了一个新的时空一致性优化模型,并基于颜色空间分布和运动空间分布特征,结合时空一致性优化方法构建了一个新的时空显著性区域检测模型。首先对视频帧进行超像素分割,然后提取三种具有互补性质的超像素级颜色空间分布特征和两种运动空间分布特征,再利用时空一致性分别融合优化空间显著特征和时间显著特征得到空间显著图和时间显著图。在时空融合阶段,利用时空一致性模型融合空间显著度和时间显著度得到超像素级的时空显著图。为进一步提高检测的准确度和完整度,通过一个能量最小化模型得到更精确的像素级时空显著图。通过与最新的视频显著性模型进行比较,本文算法有更高的准确率,对复杂背景和运动条件有强的鲁棒性。

A Spatiotemporal Saliency Framework Based on Spatial Distribution of Color and Motion

Aiming to the problem that previous spatiotemporal saliency region detection had unsatisfying performance under complex scenes and motion conditions, this paper proposed a novel spatiotemporal consistency optimization model. Meanwhile, this paper proposed a spatiotemporal saliency region detection model based on spatial distribution of color and motion using spatiotemporal consistency optimization. Firstly, it decomposed the video frame into a set of superpixels. Then, it extracted three complementary spatial distribution and two motion spatial distribution cues, then fused and optimized these saliency cues using spatiotemporal consistency optimization to derive spatial saliency map and temporal saliency map respectively. In spatiotemporal fusion phase, it derived the superpixel-level saliency map by fusing and optimizing the spatial and temporal saliency map using spatiotemporal consistency optimization. Lastly, it derived the pixel-level saliency map by solving a self-defined function energy minimization model. Compared with state of the art models, our model had better performance and stronger robust with complex scenes and motion conditions.