Extraction of textual information from images using mobile devices
In today’s world huge amount of textual data is saved in the form of photos, on paper, or as scanned documents. However, data in this form often is not readily useful as it requires a user to extract and handle data manually for any useful purposes. Particularly, sharing and storing this type of data becomes a slow and excessive ordeal for mobile users. Optical Character Recognition (OCR) techniques can be used to convert such data found in hard copies or images into editable text. Existing OCR tools are found to be excessively inaccurate in many cases as they tend to include too many wrong or misspelled words in extracted textual contents, which can cause problems to use such textual contents to use immediately for many useful purposes. Nonetheless, by applying certain preprocessing techniques on images, accuracy of an OCR output can be enhanced. In this work, we show how to improve accuracy of OCR by applying preprocessing techniques such as the top-hat transform, the Hough transform, and the affine transform on images containing textual information. Accordingly, as a demonstration, we have developed an Android app that can extract textual data from images, gather contact information such as phone numbers, email addresses, and so on. The application makes use of android intents such as phone, mail and other applications which are able to share data. Using this application, the user can get directions immediately from the current location to any address that is obtained. The text obtained can be shared to other apps in the device and also the text can be read out using text-to-speech module immediately. Our test results show that for cases where existing tools can achieve 50-60% accuracy in terms of correctly spelled words, our demonstration application can achieve 80-100% accuracy.
Hough transform, Optical character recognition, Text extraction
Agamamidi, Vijay, Dulal Kar, and Ajay Katangur. "Extraction of Textual Information from Images Using Mobile Devices." In Proceedings of the International Conference on Image Processing, Computer Vision, and Pattern Recognition (IPCV), pp. 61-66. World Congress in Computer Science, Computer Engineering and Applied Computing (WorldComp), 2017.
Proceedings of the 2017 International Conference on Image Processing, Computer Vision, and Pattern Recognition, IPCV 2017