自适应参数与边缘点引导的深度图像超分辨

Adaptive Parameters and Edge Point Guided Depth Image Super-resolution

当前TOF等深度相机仅能获取低分辨率的深度图像,无法满足三维视觉系统的需求。高分辨率深度图像可通过深度图像的超分辨算法获得,但当前算法的实验输出图像存在因纹理复制导致图像局部区域产生伪像以及边缘结构不清晰等问题。基于HR深度伪矩阵提出自适应参数与边缘点引导的深度值重建的算法,通过低分辨率深度图像边缘区域的像素点寻找修正图像中深度值错误的像素点,并进行聚类操作,从而获得重建的深度图像。通过引入自适应参数引导的自回归模型,预测边缘区域像素点的深度值。实验结果证明,该算法能够有效降低深度不连续区域的模糊性,获得更高质量的高分辨率深度图像。将该算法与当前存在算法的实验输出图像进行对比,在上采样因子分别为2、4和8时,该算法输出结果的平均坏点率均低于0.1,可有效验证算法的优势。

Depth cameras such as TOF can obtain low-resolution depth images only,and cannot meet the needs of 3 D vision systems. High-resolution depth images can be obtained by the super-resolution algorithm of depth images,but with problems such as artifacts in local areas of the image and unclear edge structures due to texture replication. Based on the HR depth pseudo-matrix,an algorithm of depth value reconstruction guided by adaptive parameters and edge points is proposed. In this paper,the pixels in the edge area of the low-resolution depth image are used to find the pixels with the wrong depth value in the correction image: and clustering is performed to obtain the reconstructed depth image. By introducing an autoregressive model guided by adaptive parameters,the depth value of pixels in the edge area is predicted. The experimental results prove that the algorithm can effectively reduce the ambiguity of the depth discontinuous area,and obtain higher quality high-resolution depth images. Compared with the existing algorithms,when the upsampling factors are 2,4 and 8,the average dead pixel rate of the output results of this algorithm is lower than 0. 1,which can effectively verify the advantages of the algorithm in this paper.