Document image super‐resolution using structural similarity and Markov random field

Abstract

Low-resolution (LR) document images may cause difficulties in reading or low recognition rates in computer vision. Thus, it is necessary to improve the resolution of an LR document image via some algorithms. In this study, a novel document image super-resolution (SR) method using structural similarity and Markov random field (MRF) is proposed. First, the non-local algorithm is utilised to find similar patches. Instead of using the Euclidian distance, a modified chi-square distance is proposed to measure the patch similarity because the bimodality characteristic of the document images can be better described by this modified chi-square distance. Finally, the structural similarity of similar patches is served as a constraint for the MRF-based SR method, which is proper to describe the neighbouring relationship between patches. The SR reconstruction for LR images of printed and handwritten documents are carried out by the proposed algorithm. Experimental results show that the reconstructed SR images obtain higher peak signal-to-noise ratio and structural similarity values than those of several state-ofthe-art SR methods and visually pleasant SR images can be produced as well.

Stroke width‐based directional total variation regularisation for document image super resolution

“…For comparison, two example‐based document image SR methods [10, 12] are experimented for document images in dataset I, using an example‐set with the same font type and size as the test document images. The total PSNR is 20.14 and 20.68 in [10, 12], respectively, while the total SSIM is 0.914 and 0.925 in [10, 12], respectively. In addition, the total OCR accuracy is 91.54 and 93.03 in [10, 12], respectively.…”

“…The total PSNR is 20.14 and 20.68 in [10, 12], respectively, while the total SSIM is 0.914 and 0.925 in [10, 12], respectively. In addition, the total OCR accuracy is 91.54 and 93.03 in [10, 12], respectively. Clearly, the results of these two methods should be the best, because other methods in Table 3 do not use any examples for SR.…”

“…The total PSNR is 17.76 and 18.69 in [10, 12], respectively. The total SSIM is 0.890 and 0.922 in [10, 12], respectively.…”

“…The bimodality regularisation term forces these grey pixels to become the background or foreground [11]. The grey levels of the background and foreground regions are obtained from the histogram of the input image. Difference with example patches : In the example‐based SR methods, an example set containing LR patches and their corresponding HR patches is utilised to generate the HR image [4, 5, 7, 9, 10, 12]. In some of these methods [4, 5], a regularisation term is used that measures the difference between the patches of the intermediate interpolated image and the extracted HR patches from the example set.…”

“…Our proposed regularisation term adaptively performs smoothing based on the local width and local direction of the character strokes, and it is called the stroke width‐based directional total variation (SWDTV). The SWDTV regularisation term can be utilised in single [4–12] or multiple [2, 3, 13–15] document‐image SR methods, as well as in example‐based [4, 5, 7, 9, 10, 12] or non‐example‐based [2, 3, 6, 8, 11, 13–15] document‐image SR methods.…”