基于MAP框架的图像序列超分辨率和模板匹配

许多图像序列处理系统的最终目的是自动或半自动地提取图像序列中的信息．为了得到被拍摄物体更清晰的影像，需要利用超分辨率技术提高分辨率．为了分析图像中的信息，常常需要利用模板匹配．以往这两个问题是分开求解的．该文提出在统一的基于最大后验概率估计的算法框架下同时求解图像序列超分辨率和模板匹配问题．算法框架具有以下两个特点：首先，在超分辨率中引入了模板库中的先验知识，提高了超分辨率的恢复效果．其次，在模板匹配中更充分地利用了图像序列中的信息，提高了模板匹配的准确性．文章的最后给出了算法在车牌识别中的实验结果．

Combined Resolution Enhancement and Pattern Matching in Image Sequence Based on MAP Frame

Many image video processing systems are designed to analysis the sequences and extract information from them. Usually, the resolution enhancement algorithms restore a high-resolution image, and then the pattern-matching algorithms find the most similar pattern to the image, and estimate the transform between them. The two steps are performed separately. This paper presents an algorithm that iteratively solves the resolution enhancement problem and the pattern-matching problem under a unified MAP frame. The frame has two advantages: first, the prior knowledge of the patterns is introduced into the resolution enhancement, and further benefits the restoration of high-resolution image. Second, compared with the former separately processing algorithm, more information of the low-resolution sequence is used in the pattern matching. The experiment and results in license recognition are presented in the paper.