Image Pre-processing for Better OCR Outputs on Business Card

Abstract

The report presents the development of business card application for accurate text recognition on Android platform. An image captured or selected from mobile phone is regards as an input of optical character recognition (OCR) and the text in the image is extracted by using OCR. Then, application classifies the output into appropriate field such as name, phone number and save it to the address book. Due to various environmental conditions in capturing image such as inadequate lighting, different perspective of taking image, therefore the accuracy of output is not satisfied. Thus, image processing techniques is used to preprocess the image before recognition. In this report, the image processing techniques used in the application will be illustrated. It includes rectification of image by finding four corners to transform image to the right position, binarization of image and removal of noise caused by lighting conditions. After implementation of image processing, the better result is return and then classifies it into different fields using algorithm and matching specified words. The error correction is executed when there is 75% similarity between output and word in database. Moreover, the application is tested with 50 business cards and gives about 80% successful rate.