CAJ | 학술논문

主动光设备是目前获取深度图的主要方法，被广泛应用于导航、人机交互、增强现实等领域。但主动光设备存在分辨率低、空洞、边缘不匹配等问题。为此，提出一种基于联合双边滤波的深度图像增强算法。采用基于深度的前景分割方法，找出深度图与彩色图边缘不匹配像素集合，利用基于联合双边滤波的插值算法对空洞进行填充。为更好保持边缘细节，增加引导深度相似项与梯度域项的方法进行插值。实验结果表明，该算法比已有方法的最小均方误差平均减少约13%，具有更好的保持边缘效果。
주동광설비시목전획취심도도적주요방법，피엄범응용우도항、인궤교호、증강현실등영역。단주동광설비존재분변솔저、공동、변연불필배등문제。위차，제출일충기우연합쌍변려파적심도도상증강산법。채용기우심도적전경분할방법，조출심도도여채색도변연불필배상소집합，이용기우연합쌍변려파적삽치산법대공동진행전충。위경호보지변연세절，증가인도심도상사항여제도역항적방법진행삽치。실험결과표명，해산법비이유방법적최소균방오차평균감소약13%，구유경호적보지변연효과。
Depth image captured by active sensor is a current tendency, which is widely used in navigation, human computer interaction, augmented reality and so on. However, common low-cost sensors have their own disadvantages, such as low resolution, holes, unmatched boundary of edge. For these problems, this paper proposes a depth image enhancement algorithm based on improved joint bilateral filtering. Depth-based foreground segmentation method is adopted to figure out the set of pixels of unmatched foreground edge, and interpolation algorithm based on joint bilateral filter is used to fill the holes. Meanwhile, in order to make further improvement, guided depth similar item and gradient item are introduced to preserve edge structure. Experimental results show that compared with the existing joint bilateral interpolation, the improved method decreases the Mean Squared Error(MSE) by 13%in average which has better effect on edge-preserving.