基于凸集投影的高分四号卫星影像超分辨率重建

基于凸集投影(projection on convex sets,POCS)对GF-4数据进行超分辨率重建处理。利用能量连续降减法构造参考帧以减少迭代次数,提高算法效率;利用Vandewalle配准法对图像块进行运动估计从而获取精确的运动参数有利于图像的分块处理。通过对GF-4真实数据进行试验与分析,表明超分图像结果的清晰度与序列影像帧数呈正相关;信噪比与帧数呈负相关。应用5帧数据进行超分重建,清晰度(平均梯度)从7.803提高到14.386;信噪比会略有下降,从3.411下降到3.336。试验表明该方法可以有效提高图像清晰度,增加图像细节信息,提升图像判别能力,扩展数据的应用范围。

GF-4 Images Super Resolution Reconstruction Based on POCS

The super resolution reconstruction of GF-4 is made by projection on convex sets (POCS).Papoulis-Gerchberg is used to construct reference frame which can reduce iteration and improve algorithm efficiency.Vandewalle is used to estimate motion parameter which is benefit to block process.Tested and analyzed by real GF-4 series images, it shows that sharpness of super resolution result is positive correlatie to frame amount, and signal to noise ratio (SNR) is negative correlate to frame amount.After processing by 5 frames, information entropy (IE) changes little;sharpness (average gradient) increases from 7.803 to 14.386;SNR reduces a little, from 3.411 to 3.336.The experiment shows that after super resolution reconstruction, sharpness and detail information of results can be greatly improved.