三维视频显著性能量计算方法研究.docVIP

PAGE PAGE 43 三维视频显著性能量计算方法研究 摘要 视觉显著性以人体视觉系统为基础，通过视觉显著性检测模型，帮助人们快速找到指定场景中的显著区域，显著性检测问题的解决能够将有限的计算机资源优先分配给显著区域，极大地提高了视频分析处理的效率对物体检测与识别、图像与视频的压缩、视频导航及质量评估等领域具有重要的应用价值。传统的基于2D的视频显著性检测方法并不能非常好地预测三维视频中的显著区域，本论文为解决立体视频中的显著性区域检测算法进行了深入研究。 本论文详细介绍了一种三维显著性模型的处理框架。通过选择无摄像机运动场景来分析一个视屏序列，然后为该场景设计显著性模型。利用自适应分割的方法进行显著性检测并在空间、时间和视差上提取显著图，最后通过计算凹和平面融合区域所占权重计算三维视频的显著性能量。 针对介绍的处理框架，本文提出了一套简单可行的针对立体视频的显著性检测框架。首先通过SUN方法提取静态显著图，通过光流法提取相邻两帧视频图像的运动显著图，最后通过基于视差的显著性检测方法提取深度显著图。通过加权求和得到立体显著图后再根据凹和平面融合区域在立体视频所占权重计算显著性能量。 最后，总结了本文课题的主要内容，并对未来进一步提高显著性检测算法的准确性做了展望。 关键词： 立体视频运动视差视频显著性显著性能量 ABSTRACT Visual saliency as the human vision system as the foundation, through visual saliency detection model help people quickly find the specified salient region in the scene, saliency detection problem solving to the limited computer resources priority in the allocation of significant regional, greatly improve the efficiency of video analysis and processing for object detection and recognition, image and video compression, video navigation system and quality evaluation field has important application value. The traditional video saliency detection method based on 2D is not very good to predict the significant area in 3D video. This paper makes a deep research on the significant region detection algorithm in stereo video. In this paper, a three-dimensional model of the processing framework is introduced in detail. A video sequence is analyzed by selecting a no camera motion scene, and then a significant model is designed for the scene. Adaptive segmentation method is used to detect and extract saliency map in space, time and disparity. Finally, the significant energy of 3D video is calculated by calculating the weight of concave and planar fusion region. According to the introduction of the processing framework, this paper presents a set of simple and feasible framework for the detection of stereo video. First of all, through the SUN method to extract the static saliency map, the optical flow method is a