结合帧选及配准的多帧超分辨率盲复原

Multiframe Blind Super Resolution Combine with Frame Selection and Registration

传统的图像复原技术不仅无法恢复截止频率之外的信息,而且没有考虑不良帧及帧间运动对复原图像质量的影响。通过对传统多帧图像盲复原模型加入运动和降采样因子进行超分辨率扩展得到一个超分辨率盲复原模型,在此基础上,首先以平均梯度和信息熵作为帧选标准,其次结合金字塔光流法对帧选后的图像进行配准,最后利用该模型进行超分辨率盲复原。实验证明,结合帧选和配准的复原方法可以抑制噪声放大及由目标空间位置偏移造成的边缘模糊。该模型是一个有效的超分辨率盲复原模型,复原图像质量优于样条插值和传统多帧盲解卷积算法。

Traditional image restoration technology not only can not restore the information out of cutoff frequency , but also can not consider the influence on the quality of the restored image with bad frame and interframe motion .Extended the traditional multiframe blind restoration model to develop a blind super resolution restoration model through join motion and down sampling factors , based on this, firstly, use the average gradient and information entropy as the standard of frame selection , secondly, use pyramid optical flow method for images registration , finally, use the model for blind super -resolution reconstruction .Experiment proves that this restora-tion method which combine with frame selection and registration can restrain the noise amplification and remove the blur caused by space position offset of the target .This model is an effective blind super resolution restoration model , the restored image quality is bet-ter than spline interpolation and traditional multiframe blind convolution algorithm .