一种固定选图数的实时幸运成像算法

A Real-time Lucky Imaging Algorithm with Fixed Number of Selected Images

幸运成像技术是一种从大量短曝光图像中选取少量幸运好图进行配准、叠加的高分辨率图像恢复技术,能够有效减小大气湍流对图像质量的影响,但传统的基于中央处理器(Central Processing Unit,CPU)的幸运成像算法难以实现实时化。利用现场可编程门阵列(Field Programmable Gate Array,FPGA)的并行性和灵活性优势,提出了一种新的基于现场可编程门阵列的幸运成像算法,并构建了一个现场可编程门阵列实验系统。该算法采用一种固定选图数且无需排序的图像选择策略和一种以行列坐标为基准的图像配准策略,能够有效节省算法处理时间和硬件资源,达到实时幸运成像的目的。实时幸运成像算法能够以简洁的方式在中小规模的现场可编程门阵列上实现,所得高分辨率图像与基于传统中央处理器算法处理的结果完全相同。实验表明,对于2000帧128×128像素的输入图像进行幸运成像,本算法的运行速度比本实验室之前提出的算法快27倍,比传统的基于CPU+MATLAB幸运成像算法速度快150多倍,处理帧率可达197帧/秒。该算法及其现场可编程门阵列实现技术可以用于构建真正实时的幸运成像系统。

Lucky imaging is a high-resolution image restoration technology by selecting,registering and superposing a small number of lucky images from a large number of short-exposure images.It can effectively reduce the effect of atmospheric turbulence on image quality.However,the traditional CPU-based lucky imaging algorithm is difficult to achieve real-time.Taking advantage of the parallelism and flexibility of FPGA,this paper proposes a novel FPGA-based lucky imaging algorithm and built an FPGA experimental system.The algorithm adopts an image selection strategy only selecting a fixed number of lucky images and without sorting,and an image registration strategy based on the row and column coordinates of the selected images,which can effectively save algorithm processing time and hardware resources,and achieve the purpose of real-time lucky imaging.The new real-time lucky imaging algorithm can be implemented on small and medium-sized FPGA in a simple way.The high resolution image obtained by the real-time lucky imaging algorithmis the same as that processed by the traditional CPU-based algorithm.Experiments show that,for a lucky imaging processing of 2000 frames of the input images of 128×128 pixels,the running speed of the FPGA-based algorithm is 27 times faster than that of the algorithm previously proposed by our laboratory,and over 150 times faster than that of the traditional lucking imaging algorithm based on CPU+MATLAB,the processing frame rate is up to 197 fps.The proposed algorithm and its FPGA implementation technology can be used to build a true real-time lucky imaging system.