Pre-Processing of Degraded Printed Documents by Non-local Means and Total Variation

Abstract

We compare in this study two image restoration approaches for the pre-processing of printed documents: namely the Non-local Means filter and a total variation minimization approach. We apply these two approaches to printed document sets from various periods, and we evaluate their effectiveness through character recognition performance using an open source OCR. Our results show that for each document set, one or both pre-processing methods improve character recognition accuracy over recognition without preprocessing. Higher accuracies are obtained with Non-local Means when characters have a low level of degradation since they can be restored by similar neighboring parts of nondegraded characters. The Total Variation approach is more effective when characters are highly degraded and can only be restored through modeling instead of using neighboring data.

Enhancement of historical printed document images by combining Total Variation regularization and Non-local Means filtering

“…TV and NLmeans can be used as single enhancement approaches and applied directly to document images [25]. We applied these methods in isolation at the ICDAR DIBCO 2009 document image binarization contest [18] and our results were reported top 10 (over 43 participants) according to the four measures used in the contest, and top 6 according to the F-measure only.…”

“…But since NLmeans never completely discards patches, this tends to smooth too much and we have shown in [25] that fading characters lose more pixels when the search window is too wide.…”

“…It can be noted that preliminary experiments on set XX-b [25] used a common global threshold for XX-b images, the value (75) being chosen as suggested in [36] for this data set. However for the sake of comparison, all sets are binarized in our experiments according to the same Otsu-based framework.…”

Effect of pre-processing on binarization

“…The use of Total Variation (TV) as a pre-processing filter was found to improve OCR results on an OCR system that applied an Otsu threshold to all incoming images. 7 Thus its comparison with other pre-processing filters for image binarization was initiated. The Total Variational formulation of the image restoration problem consists of minimizing a weighted combination of the data fidelity and the regularization (also called prior).…”

A mask-based enhancement method for historical documents

“…TV has one main parameter f3 which determines the balance between the data fidelity term and the TV term. 19 We use a default value f3 = 20.…”