基于三维变换域频谱差的视频显著性检测算法

显著性区域检测是计算机视觉的重要课题,对视频质量评价和感知视频编码算法优化也至关重要.大多显著性检测算法不能权衡准确率和复杂度,限制了它们在视频预处理和实时处理中的应用.提出了一种基于三维变换域频谱差(3DTDSD)的快速视频显著性检测算法,分别以关键帧及其前一帧为中心建立一定数量图像帧的滑动窗,得到两组3D视频体,用傅里叶变换将两组视频变换到三维频域,两组三维数据之间的差值经过反变换得到显著性图,最后通过连通分析、阈值判断等得到显著区域.频域算法具有运算速度快的特点,实验对比和算法复杂度分析证明了该算法的有效性和快速性.

Video saliency detection algorithm based on 3D transform domain spectral difference

Saliency region is very important for video quality assessment and the optimization of perception video coding algorithm, it is also an important issue in computer vision. However, the accuracy and complexity of most algorithmes are not uniform well, which limites their application in the video preprocessing and real-time processing. In this paper, a fast video saliency detection algorithm based on 3D transform domain spectral difference (3DTDSD) is proposed. First, it takes the keyframe and its previous frame as the center, establishe a sliding window to get two sets of 3D video cube; Second, with a Fourier transform, the cubes are transformed into three-dimensional frequency domain; Third, the difference between the two sets of three-dimensional data after inverse transform can get the saliency map. Finally, through the connectivity analysis and threshold judgement will get the saliency region. The experimental results show that the algorithm improve the accuracy of the test results, and computational complexity is also significantly reduced .