基于DSP系统的超分辨率图像重建技术研究

Super-Resolution Image Reconstruction Technology Based on DSP System

由于航空光电设备造价与体积等的限制,需要在不改变航空光电设备硬件结构的前提下,获取尽可能清晰的图像或视频。文章提出了基于DSP图像处理系统的超分辨率重建方法,首先利用Fourier-Mellin变换法和Keren算法的联合优化算法进行运动估计;然后利用基于边缘保持的凸集投影简化方法进行超分辨率重建;最终结合DM642的特征,在不降低精度的前提下,对算法进行优化实现。该方法在不增加系统结构体积和成本的前提下,有效地提高了成像系统的分辨力,进而提高系统的目标识别能力。在以DM642为核心嵌入式图像处理平台中实现超分辨率重建实验,所采用的相机分辨率为720×576,整个重建的时间由传统的几分钟甚至几十分钟下降至20s左右。实验结果表明,用本文方法重建出的图像细节明显比单帧插值的图像清晰,图像的平均梯度和信息熵有了明显提高。

Because of the limitations of cost and size of the navigation photo-electricity equipments,it is necessary to get images or videos as clear as possible without changing the structure of navigation photo-electricity equipments.This paper introduces a method based on DSP image processing system to realize super-resolution reconstruction which realizes motion estimation by the united optimized algorithm of Fourier-Mellin and Keren firstly,and achieves super-resolution image reconstruction based on projection onto convex sets,lastly,realizes the method on DSP by optimizes the method based on the feature of DM642 without reducing the accuracy.This method improves the resolving power of the image system without increasing the size of the system structure as well as enhance the identify ability of the system.This paper has proposed to realize super-resolution reconstruction on the embedded image processing platform which the core is DM642,the resolution of camera is 720×576,and the reconstruction time decreases to 20 s from several minutes or even several ten minutes.Experiment results indicate that the detail of reconstruction image is clearer than that of the single frame interpolation image,and the average gradient and the information entropy also get improved.