一种基于字典学习的深度图像超分辨率重建方法

Method of Depth Image Super-Resolution based on Dictionary Learning

随着深度图像获取技术的快速发展,深度信息(拍摄物与成像平面的距离)已广泛应用在人机交互、虚拟现实和三维重建等领域。然而,由于成像分辨率低和成像质量差等因素影响,传统深度相机(如TOF,Time-of-flight)无法很好满足实际应用需求。研究深度图像的超分辨率重构问题,旨在提高深度图像的分辨率与质量,提出一种基于字典学习的深度图像超分辨率重建方法,首先从样本中获取先验信息,采用学习策略对低分辨率图像进行初始深度重建;然后获得足够的边缘信息,利用边缘信息对初始重建图像进行联合重建,最终获得高分辨率深度图像。实验结果表明,相比传统方法,本文方法能够取得更好的深度图像超分辨重建效果。

Recently, depth images, which captured by depth cameras such as TOF（time-of-flight）, with pixel values representing the distances between objects and the imaging plane, are widely used in virtual reality, man-machine interaction, 3-D scenes reconstruction and other fields. However, these applications are limited by the low resolution and low quality of the depth images. Address the super-resolution of depth image, aiming at the improvement of both resolution and quality. A super-resolution method which adopts the dictionary learning for depth image is proposed. Firstly obtain the sufficient prior information from the sampled depth images to recovery an initial high resolution image; Then, sufficient edge information is obtained for a second reconstruction combining the initial reconstruction to obtain the higher performance. The experimental results show that the proposed method achieves better performance than the referenced.