基于保边滤波的显著目标快速分割方法

Fast Salient Object Segmentation Method Based on Edge-Preserving Filtering

在视频中自动发掘目标并对其进行精确分割是一个非常有挑战性的计算机视觉问题。本文提出了一种基于保边滤波的显著目标快速分割方法。首先,通过融合外观特征与运动特征,将视频中的显著目标发掘转为能量函数最小化问题进行求解。其次,为了更精确地进行分割目标,融合外观的高斯混合外观模型(Gaussian mixture mode,GMM)、位置先验以及时空平滑约束构建马尔科夫随机场(Markov random field,MRF)模型,并使用图割算法进行求解。本文提出的基于保边滤波的显著目标快速分割方法,在牺牲较少的精度下,极大地提高了分割效率。最后在两个数据集上进行了对比实验,实验结果表明,本文算法的分割精度超过了其他5种目标分割方法,且加速算法在损失少量精度的情况下提高了2倍分割效率。

How to automatically discover salient objects in video and further perform accurate object seg- mentation is a challenging problem in computer vision. Here, fast salient obiect segmentation method based on edge-preserving filtering is proposed. Firstly, the salient object discovery is formulated as an en- ergy minimization problem, which fuses the appearance and motion features. Then, a Markov random field （MRF） model, integrating the Gaussian mixture model （GMM） of appearance, the location prior, and the spatial-temporal smoothness, is constructed for accurate segmentation, and is efficiently opti- mized by graph cut. Moreover, an edge-preserving-based method is presented to improve the segmenta- tion efficiency with a little loss of accuracy. Finally, extensive experiments on two datasets suggest that the proposed method performance is better than that of other five methods, and the accelerated version can speed up to two times of the original one.