Abstract: The prevalence of handheld devices such as mobile phones for image capturing now a days is uncontested which is mainly contributed by their high image quality, low cost and portability. Hence, it is only natural that users would prefer the convenience of documents image capturing to photocopying or scanning. Images taken from these devices however, usually succumb to uneven illuminations of which photocopied or scanned documents are spared. This study expounds a simple thresholding technique to eliminate unwanted background of document images captured by handheld devices. The technique administers a local thresholding which is a simple computation of the mean and standard deviation of sliding windows procured from the whole image. We appraise the efficacy of method visually and analytically. In analytical experiments we use 5 uneven illuminated document images captured from a 5MP mobile phone camera and 13 images of DIBCO. Results of several performance measurements from the technique is compared with those generated from several state-of-the-art thresholding methods. The results attest to the efficacy of the Otsu technique in eliminating background, albeit in the absence of uneven-illumination and a large gap between image pixels contrast. In addition, the techniques propounded by Sauvola, Nick and Bataineh perform well with uneven illumination cases, but fall short in low contrast and when text pixels values are too close to the foreground or background pixels values. Unlike the proposed technique, which perform well in both circumstances. This substantiates the claim that performance of the proposed method is superior to that of the other techniques in overcoming uneven illumination especially on shadow cases and close value proximity of the text with that of foreground or background as well with low contrast.
Musab Kasim Alqudah, Mohammad Nasrudin, Bilal Bataineh and Khairuddin Omar, 2016. A Novel Simple Thresholding for Uneven Illuminated Document Images Captured via Handheld Devices. Asian Journal of Information Technology, 15: 2927-2936.