改进的暗通道去雾算法在多核DSP上的并行实现

Parallel Implementation of Improved Dark Channel Defogging Algorithm on Multi-Cores DSP

为提高雾天拍摄图像的质量,本文在暗通道先验去雾算法的基础上,提出了一种改进算法.首先,用引导滤波替代软抠图对透射率进行优化,降低了算法的复杂度,易于在多核TMS320C6678上并行实现;其次,对明亮区域的透射率进行优化使复原图像的色彩失真得到明显改善.改进算法无论是对自然光图像,还是遥感图像都有相当好的去雾效果.对一系列含雾图像的实验表明,改进算法在TMS320C6678上八核并行实现,当DSP内核工作主频为1GHz时,单幅像素为600×800、1 200×1 600、1 800×2 400的图像的运行时间分别为38.21、154.22、347.18ms,并行效率最高达7.35,满足实时复原含雾图像的要求,提高了图像的呈现质量,为视频去雾处理打下了良好基础.

In order to improve the quality of the images in fog days,an improved algorithm is proposed based on the dark channel prior defogging algorithm.Firstly,optimize the transmission with the bootstrap filter instead of soft matting,reducing the complexity of the algorithm;Secondly,the optimization of the transmission on the bright region has been significantly improved the color distortion of the restoration image.This improved algorithm has a good effect on the natural light images and the remote sensing images.The improved algorithm is successful parallel operating in the TMS320C6678 platform with eight cores after a series of verification experiments on fog images.The DSP core sets its dominate frequency at 1GHz and the running time is 38.21,154.22 and 347.18 ms in the case that the pixels of images are 600×800,1 200×1 600 and 1 800×2 400,respectively.The parallel efficiency is up to 7.35 at this moment.The implementation of algorithm in this paper which improved the quality of images and laid a good foundation for video defogging.