基于半像素运动信息的快速超分辨率图像重构算法

A Fast Algorithm Based on Sub-Pixel Motion for Super-Resolution Reconstruction of Image

提出了一种全新的、利用半像素运动信息进行超分辨率重构的快速算法.该算法连续两次使用了图像的一维统计特性:第一次将图像分为运动区域和背景区域;第二次针对运动区域估计各像素的运动范围.在对运动区域进行运动估计和运动补偿后获得N个运动补偿图,最后结合背景区域,利用数据融合技术获得超分辨率的重构图像.试验结果表明,该算法计算量小、复杂度低,无论是最小均方误差,还是信噪比或峰值信噪比,均好于传统的双线性内插法.

A novel algorithm based on sub-pixel motion for super-resolution reconstruction of images is proposed. In this algorithm, the 1-dimension statistical characteristic of the image is used twice. Firstly, the N frames of the image are divided into the motion field and the background field. Secondly, the motion-range of the pixels in the motion field is estimated. After the motion estimation and motion compensation, N motion compensation images are obtained. At last, the data fusion technology is adopted to reconstruct a super-resolution image in combination of the information in the background. The experimental results show that the proposed algorithm is simper and less complex. The performance of this algorithm is better than the traditional bilinear interpolation method in minimum mean square error, signal-to-noise ratio and power signal-to-noise ratio.