A SIFT-Based Approach of Recognition of Remotely Mobile Phone Captured Text Images

Mobile phones have been the trending and indispensable devices to many people these days. With the advancement of science and technology, inexpensive and versatile phones can be easily found at one's disposal. The field of Optical Character Recognition (OCR) is also promoted, as OCR mobile applications are increasingly chosen. This paper addresses an approach to solve one challenging issue of OCR, which is, recognition of text images taken by mobile phones at remote distances. These images usually appear in small font size and at low resolution, as a result, yield very poor outcomes when fed to OCR systems. The remote text images are first analyzed using the vertical projection profile. Then the words and characters are extracted, respectively, after determining the location of peaks and space in the graph. Finally, SIFT algorithm is applied in recognition stage, using the low resolution text templates collected, to find the best match for the texts. The results showed an improved performance in OCR recognition rate, compared to OCR reading software.