中文摘要：

针对序列图像超分辨率重建非局部均值（non-local means,NLM）算法重建结果图像边缘区域过平滑的问题,提出了一种局部参数自适应改进方法。将整幅图像划分为图像子块,然后根据图像子块平均像素信息计算出其对应的滤波参数,这样有助于减少因整幅图像使用统一滤波参数而导致的某些高频信息的丢失。实验结果表明,与经典NLM重构算法相比,改进算法重建出的结果图像的轮廓边缘更清晰,字符辨识度更高;在算法实现方面,图像重构程序在CPU/GPU平台上实现,使用GPU并行化加速的程序比单CPU运算的程序,加速比最高可达到30倍,显著缩短了重构程序计算时间,提高了该图像超分辨率重建算法应用于实际场所的可能性。

英文摘要：

For non-local means algorithm of super reconstruction using image sequences likely leading to smoothing the edges of reconstructed image, this paper proposed an improvement of local parameter adaptive. First of all, it divided the whole ima- ge into sub-blocks. Then according to the average pixel information, calculated the corresponding filtering parameter for each sub-block, which help to reduce the loss of high frequency information due to the whole image using unified filtering parame- ter. The result demonstrates that this modified algorithm significantly improves the clarity of edges and the recognition rate of character in reconstructed image compared to the classical NLM algorithm. It is found that the processing time on GPU is much less than on CPU, and the highest speedup ratio to the traditional algorithm is more than 30 times. It raises the possibility that applying this super-resolution algorithm into the actual workplace.