一种基于实例的文本图像超分辨率重建算法

Algorithm for document image super-resolution reconstruction based on examples

为了从一幅包含文字、公式和图形等内容的低分辨率文本图像重建高分辨率图像,提出了一种获取重建图像先验知识的新方法.利用实例图像和图像降质模型建立图像库,图像重建时,将低分辨率观测图像分成若干子块,每个子块分别从图像库中找到一块最佳匹配的高分辨率实例图像块,将这些实例图像块依次拼成一幅大图,并把该大图各点的灰度值作为重建图像各点灰度值的均值,以此先验知识采用最大后验概率(MAP)准则估计出高分辨率文本图像.实验结果表明本文的方法能够取得较好的重建效果.

In order to produce a high-resolution image from a low-resolution document image containing characters, equations and graphics, a new method to obtain the prior knowledge of the high-resolution image is proposed. Image examples and degradation model are used to generate example im- age database. A high-quality patch is assigned to each block in the observed low-resolution image, whose corresponding low-quality patch is found as the nearest neighbor in the image database. These high-quality patches are mosaicked to produce an enlarged image whose pixel intensities are taken as the mean values of the pixel intensities of the desired high-resolution image. A maximum a posteriori （MAP） estimator is used to estimate the high-resolution image. Experimental results show that the new method improves the reconstruction results significantly.