基于C6416的匀速直线运动的模糊图像恢复

Real-time restoration of images blurred by uniform linear motion based on TMS320C6416

为解决相机与景物之间由于相对运动而造成的图像模糊问题,提出在定点DSPC6416上,通过优化维纳滤波算法来实现模糊图像的实时恢复。提出基于实数的一维维纳滤波优化算法。给出该算法与其他常规维纳滤波算法的实验结果和对比评价。实验结果显示:在主频600MHZ的DSPC6416实验平台上,对512×5128位灰度测试图片先进行模糊处理,再通过该优化算法恢复,速度达到58ms每帧,而常规一维维纳滤波算法速度为90ms每帧,二维维纳滤波算法速度为253ms每帧。恢复质量三者相同。结果表明该算法每秒能处理17帧512×5128位灰度图片,基本能满足实时性的要求。如果采用最新的主频1GHZ的DSPC6416芯片,处理速度预计能达到35ms每帧,能实现25帧每秒的实时性要求。

To absolve image blur phenomena caused by relative motion between camera and objects,a real-time blur image restoration method is put forward, which contains the technique of optimum Wiener filter based on the fixed point DSP C6416.A new one dimension Wiener filter based on real number is brought forward and compared with other Wiener filters about the results of the contraposed experi- ments. On the fiat of DSP C6416 with the frequency of 600MHZ,an image of 512×512 8 bit was blured first to provide the fuzzy one which was resumed by the optimum Wiener filter and cost 58ms per frame at the second step. The general one dimension Wiener filter need 90ms per frame and the two dimensions one need 253ms per frame. But the qualities of the resumed images are the same about the three methods. The results demonstrate that a 512×512 8 bit blured image can be resumed at the speed of 17 frames per second by the optimum Wiener filter, so the method satisfy the real time application. Supposed that the newly DSP C6416 with frequency of 1GHZ was used, the speed of it can reach 35ms per frame so it can arrive at the real time guideline of 25 frames per second.