本文提出了一种基于支撑向量机的盲超分辨率图像复原算法.首先采用Sobel算子和局部方差从训练图像中提取能够表征模糊参数信息的特征向量,并利用支撑向量机建立特征向量与对应的候选参数的映射关系,然后通过建立的模型对不同光照条件下的低分辨率图像进行参数辨识,最后根据辨识出的模糊参数融合不同光照条件下的低分辨率图像同时实现了图像动态范围和空间分辨率的增强.为了实现低分辨率图像间的亚像素配准,还提出了一种基于Retinex的亚像素运动估计算法.仿真结果表明与传统算法相比,无论从主观视觉还是定量描述上本文算法均具有较好的效果.

In this paper we propose a blind super-resolution image restoration algorithm based on Support Vector Machines (SVM).Firstly,Sobel operator and local variance were used to extract feature vectors that contain information about different Point Spread Functions (PSFs) and SVM was used to classify these feature vectors.The acquired mapping between the vectors and corresponding blur parameters provided the identification of the blur.After blur identification,a super-resolution image was reconstructed from several low-resolution images obtained in different illumination conditions.The reconstructed image has high spatial resolution and dynamic range.We also propose a sub-pixel registration algorithm based on Retinex theory.Simulation results demonstrate the effectiveness and higher performance of the proposed method in both objective measurements and subjective visual quality.