针对现有光场深度重建算法只能获取单一视角深度信息,三维建模应用受限的问题,提出一种大视野图像拼接算法。该算法基于三维光场重建的深度图像序列,采用双边滤波及插值算法对图像进行去除噪声,利用基于边缘曲率极值的角点检测算法,并通过最小曲率偏差的列间度量匹配实现了图像之间的精确配准,采用改进的加权平均方法对重叠区域进行融合操作,既保留图像细节又拓展图像视野。实验结果表明,该算法能够有效实现光场深度图像序列的平滑拼接,有利于形成大视野深度数据,为大场景的三维重建奠定了基础。

A new image stitching algorithm is proposed based on depth images series, to overcome the limit of field-of-view of depth reconstruction from three-dimensional light field. At pre-processing stage, bilateral filter and interpolation is utilized to de-noise. A corner detection algorithm based on edge local-maximum curvature value, and a matching algorithm based on minimum curvature variance control are proposed. Improved weighted average method is adopted to complete image fusion, both retain image's details and expand field-of-view. Experimental results show that the proposed system can easily obtain a complete depth image, which is helpful for three-dimensional reconstruction of whole scene.