Image Preprocessing For Geometric Feature Extraction in OCR Systems

Optical character recognition (OCR) is one of the most successful application of pattern recognition and image processing. Character geometry is one of the most useful feature for identifying characters in images. The geometric feature extraction techniques proposed in literature are complex and requires extensive effort in implementation. In this paper, we propose a preprocessing technique which is simple and makes the feature extraction process easy. The proposed system is intended for preprocessing colour images containing printed English alphabets. The image is fed to a preprocessing unit where it is first converted to gray and then thresholded to obtain a binary image. Next step is to filter the noise present in the image. Now the area containing text will be identified and segmented. Each character is extracted from the segmented area in the image and apply morphological thinning operation to make the feature extraction process easy. Keywords— Geometric feature extraction, Optical character recognition, Preprocessing, Structural analysis.