面向视差图优化的可信性三边滤波算法  

Confident Tri-lateral Filter for Disparity Map Refinement

作　　者：朱长水[1] 袁红星[2] 袁宝华[1] 曹红根[1] 

机构地区：[1]南京理工大学泰州科技学院,江苏泰州225300 [2]宁波工程学院电子与信息工程学院,浙江宁波315016

出　　处：《小型微型计算机系统》2015年第12期2773-2777,共5页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61101197/F010402)资助

摘　　要：视差图的不连续过渡是决定3D体验的关键指标.现有以边界保持滤波为基础的增强方法没有考虑到滤波器核中来自不同对象像素点的影响,导致增强的视差图在边缘处出现不同程度的失真或模糊.为此,提出一种利用视差图可信性的三边滤波方法.首先统计视差图的变化差异,将视差变化较大的像素点分配较低的可信度;其次,由可信度、空间相邻性和颜色相似性构建三边滤波核;最后,用该滤波核对视差图进行滤波增强.美国Middlebury大学立体视觉算法在线评测系统的实验数据表明,该方法能够较大提升原始视差图的性能指标,定性和定量指标均优于现有的双边滤波、联合双边滤波和三边滤波算法.Discontinuities of disparity map are very important for creating realistic looking depth effects. The existed disparity disconti- nuities enhancement methods based on edge preserving filter will introduce distortion or blur to the object boundary as impacts of sup- port pixels on another object are not considered. So, a tri-lateral filter based on confidence is provided. Firstly, low confidence is as- signed to pixels which have the larger disparity changes. Then, a tri-lateral filter kernel is set with spatial weight, color weight and con- fidence factor. The enhanced disparity map is filtered with the kernel at last. The experimental data from the Middlebury stereo online evaluation system shows that our method can improve the original disparity map performance greatly and it also performs better than bilateral filter ,joint bilateral filter and tri-lateral filter.

关 键 词：立体匹配 视差 双边滤波 三边滤波 可信度 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]