A Morphological Image Preprocessing Suite for OCR on Natural Scene Images

As demand grows for mobile applications, research in optical character recognition (OCR), a technology well-developed for document imaging, is shifting focus to the recognition of text embedded in digital photographs or video. Segmenting text and background in natural scenes is a difficult classification problem, and the accuracy of this segmentation is of utmost importance when the output of an OCR system will be transformed as in translation or speech synthesis. Our work proposes an image preprocessing suite that, through text detection, auto-rotation, and noise reduction, improves the accuracy of OCR analysis in a camera-based translation system. Our novel approaches for foreground/background detection and skew estimation using morphological edge analysis show immediate improvement in OCR accuracy, and our experimental classification of text regions using unsupervised feature-based clustering provides a good background for future research in applying signal processing and machine learning techniques to the problem of text detection.