Deep Learning Approaches to Determining Optimal Resolution for Scanned Text Documents

Abstract

Image quality assessment has been a very active research area in the field of image processing, and there have been numerous methods proposed. However, most of the existing methods focus on digital images that only or mainly contain pictures or photos taken by digital cameras. Traditional approaches evaluate an input image as a whole and try to estimate a quality score for the image, in order to give viewers an idea of how “good” the image looks. In this paper, we mainly focus on the quality evaluation of contents of symbols like texts. Judging the quality for this kind of information can be based on whether or not it is readable by a human, or recognizable by a decoder such as an OCR engine. We mainly study the quality of scanned documents in terms of the detection accuracy of its OCR-transcribed version. For this purpose, we proposed a novel CNN based model to predict the quality level of scanned documents or regions in scanned documents. Experimental results evaluated on our testing dataset demonstrate the effectiveness and efficiency of our method both qualitatively and quantitatively.