一种使用边缘方向图的去马赛克算法

A Novel Demosaicing Algorithm with Edge Orientation Map

提出一种适用于Bayer格式图像的去马赛克算法,其核心在于为重建插值产生准确的边缘方向图.首先,由改进的边缘检测算子产生初始的边缘方向图,并使用十字形中值滤波器对初始的边缘方向图进行滤波,最大程度地去除初始方向图中孤立的错误插值方向.其次,基于滤波后的边缘方向图,运用统计分析以确定细致纹理及边界交界处的插值方向.最终,对在准确的边缘方向图的指导下完成重建插值的图像进行细化以消除插值人工痕迹.实验证明,本算法不仅颜色峰值信噪比(color peak signal-to-noise ratio,CPSNR)高于其他几种经典及近期算法,在边缘细致区域依然保持了很高的恢复质量,而且保持了较低的计算量,在性能和开销之间取得了很好的平衡.

In this paper, we present a novel cost effective demosaicing algorithm for Bayer images. In the proposed method, improved direction detectors are firstly derived from the CFA information to produce an initial edge orientation map. Then the initial edge orien- tation map is passed through a cross-shape median filter to remove most isolated wrong directions. A statistical analysis is also per- formed in the filtered edge orientation map to decide the interpolation directions for pixels locates in the regions of fine texture or at the borders of adjacent edges. Further, a refine step is included to reduce the interpolation color artifacts. Experimental results con- firm that the proposed algorithm performs better compared to several classic and concurrent techniques, both subjectively and objec- tively. Besides, the proposed method provides a good compromise between image quality and computational cost.