基于模式分类和图像融合的超分辨率方法

Super-resolution based on pattern classifying and image fusion

在低分辨率图像序列的超分辨率重建过程中,如何由配准后的视频序列构造出高分辨率图像对重建结果起着至关重要的作用,而现有算法只是采取了求均值的方法,这就削弱了细节信息。新算法根据运动估计的位移对低分辨率序列进行分类,在各类内根据重叠区域再进行分类,然后采用基于方向信息测度的方法进行数据融合,最后输出高分辨率图像。试验表明提出的算法简单、有效,增强了超分辨率算法的信息搜集能力。

In the process of super-resolution, how to combine registered images is a very important step. But existing methods pay little attention to it, which just takes mean of images. As a result, details are smoothed. According to translation component estimated and overlapped area, the proposed algorithm first classified preprocessed images, then in each class it performed image fusion based on orientation information measure, and finally it output high resolution image. Experimental results show that the proposed algorithm is simple and effective and it enhances the ability of super resolution algorithms to extract details from observations.