基于多尺度边缘保持正则化的超分辨率复原

Super-Resolution Restoration with Multi-Scale Edge-Preserving Regularization

超分辨率复原是一种由一序列低分辨率变形图像来估计一幅(或一序列)较高分辨率的非变形图像的技术,同时,它能够消除加性噪声以及由有限检测器尺寸和光学元件产生的模糊.提出了一种基于多尺度正则化先验的最大后验概率超分辨率复原算法.算法特点如下:(1) 对运动估计结果实施可信度验证;(2) 采用图像的多尺度小波表征来定义图像的空域活动性测度,并由此构建多尺度Huber-Markov先验模型.实验结果表明,该算法不仅具有较好的超分辨率图像边缘保持能力,而且能够有效地消除图像伪迹.该算法可以应用于遥感图像、医学成像、高清晰度电视标准和合成视频变焦等领域.

Super-Resolution restoration is a technique for estimating an unaliased high-resolution image (or a sequence) from an aliased video sequence and combating additive noise and blurring due to the finite detector size and optics. An improved Bayesian MAP estimator with multi-scale edge-preserving regularization for super-resolution restoration is proposed. The confidence of the motion estimation result is validated to eliminate motion artifact. The wavelet representation of an image is utilized to define the spatial activity measure of the image, and further to construct a novel multi-scale Huber-Markov model. The experimental results show that the multi-scale Huber-Markov model can be incorporated into Bayesian MAP estimator to preserve the edges of the super-resolution image effectively. This proposed algorithm is widely used for remote sensing, medical imaging, high-definition television (HDTV) standard and creation of synthetic 'video zoom'.