Text detection and recognition using enhanced MSER detection and a novel OCR technique

Detection and recognition of text from any natural scene image is challenging but essential extensively for extracting information from the image. In this paper, we propose an accurate and effective algorithm for detecting enhanced Maximally Stable Extremal Regions (MSERs) as main character candidates and these character candidates are filtered by stroke width variation for removing regions where the stroke width exhibits too much variation. For the detection of text regions, firstly some preprocessing is applied to the natural image and then after detecting MSERs, an intersection of canny edge and MSER region is produced to locate regions that are even more likely to belong to text. Finally, the selected text region is taken as an input of a novel Optical Character Recognition (OCR) technique to make the text editable and usable. The evaluation results substantiates 77.47% of the f-measure on the ICDAR 2011 dataset which is better than the previous performance 76.22%.