改进的Bayer插值算法及其硬件实现  

Improved Bayer-pattern demosaicking and its hardware design

作　　者：朱波[1,2] 汶德胜[1] 王飞[3] 

机构地区：[1]中国科学院西安光学精密机械研究所,陕西西安710119 [2]中国科学院研究生院,北京100049 [3]西安交通大学,陕西西安710049

出　　处：《光电子．激光》2013年第6期1211-1218,共8页Journal of Optoelectronics·Laser

基　　金：国家自然科学基金(90920301)重点资助项目

摘　　要：为了改善彩色恢复效果,降低硬件设计的复杂度,提出了一种基于Bayer格式图像彩色插值改进算法。首先,对Bayer格式原图进行二阶拉普拉斯修正,获取水平方向和垂直方向的绿色分量^GH和^GV;然后,计算红色通道和蓝色通道分别与绿色分量的水平色差R-^GH、B-^GH和垂直色差R-^GV、B-^GV,再对色差方向插值,得到水平和垂直方向的梯度并进行比较,根据比较结果选择^GH或^GV作为选定像素处的绿色分量^G;当整幅图像的绿色分量得到后,利用给定像素周围的4个色差进行插值计算,分别得到红色分量^R和蓝色分量^B,从而完成彩色图像的恢复。试验结果表明,改进的算法在峰值信噪比(PSNR)上相比双线性插值(BI)算法,R、G、B三通道均提高15dB左右,且图像目视细节丰富,伪彩色得到有效抑制,特别是硬件实现难度方面和线性算法相当,能够做到硬件实时处理。目前,算法已经在Xilinx FPGA上实现,并工程应用,满足了项目实际需要。In order to improve the quality of demosaicked images and reduce the complexity of its hardware design, an improved Bayer-pattern demosaicking method is presented. Firstly, the second-order Laplacian interpolation filter is adopted tO interpolate the missing green samples G^H and G^v at red and blue pixels. And then,in every red or blue location, we calculate the horizontal chrominance values R-G^H, B-G^H and vertical ones R--G^v, B-G^v. Next, we calculate and compare the horizontal chrominance gradient and the vertical one for choosing the missing G. After the green channel has been reconstructed,we interpolate R and B as the average of the four nearest neighbors. By now,the full color image is reconstructed. Experimental results prove that on average the improvement is 15 dB over the bilinear algorithm （BI） in peak signal-to-noise ratio （PSNR） and it appears to produce visually more pleasant color image details with color artifacts greatly suppressed. Furthermore, the computational cost of the proposed algorithm is as simple as the linear algorithm, so the interpolation can be hardware processed in real time. Now, the proposed algorithm is implemented with Xilinx FPGA and used in our camera system, and the perfect restoration quality and effectiveness can meet the requirement of system design.

关 键 词：BAYER格式 彩色插值 梯度 FPGA 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]