基于超像素与自回归模型的深度恢复  

Depth Recovery Based on Super-pixels and a Color-guided Auto-regressive Model

作　　者：孙鑫[1] 杨敬钰[1] 姜斌[1] 李坤[2] 

机构地区：[1]天津大学电子信息工程学院,天津300072 [2]天津大学计算机科学与技术学院,天津300072

出　　处：《光电子技术》2014年第2期121-124,共4页Optoelectronic Technology

基　　金：国家自然科学基金(61072062)资助项目

摘　　要：提出基于超像素分割,并联合自回归模型的深度复原方法。首先对已获取的场景彩色图进行过分割,得到彩色图中每个像素的标号,然后构建基于已分割彩色图像指导的自回归模型:根据像素标号对自回归模型系数进行预测,通过优化预测系数差错来实现深度图恢复。实验表明,该算法不仅能有效恢复出深度图,而且在结构边缘细节处更加突出,优于目前主流的方法。A new depth image recovery method which is based on super-pixels and the adaptive color-guided auto-regressive(AR)model is proposed.First,the accompanied high quality color image is segmented,with the label of each pixel is obtained.Then,an adaptive segmentedcolor-guided auto-regressive(AR)model is constructed:the AR predictor is constructed according to the label of each pixel.By minimizing AR prediction errors,the recovered depth image is obtained.It is experimentally proved that our method not only recovers the depth image effectively but also overpasses current mainstream methods for thin structures or sharp discontinuities.

关 键 词：深度图恢复 超像素 自回归模型 系数预测 深度相机 

分 类 号：TN919.82[电子电信—通信与信息系统]