Automatic text detection and digital character recognition

Text and optical character recognition (OCR) plays an increasingly important role in many modern applications in the field of medicine, finance, transport, and security. While many systems have been designed to identify text within an image, those methods were designed to deal with special types of images, such as those from scanned documentsý, web page, and newspapers. With variability in image quality, text format, and varying degrees of resolution, accurately extracting data embedded in an image is an open research problem which inspired researchers to focus on text retrieval techniques. In addition, the recognition of degraded text printed on a clear plastic surface has not been addressed.

This research provides a solution to the problem of text extraction from images captured by a visual sensor. This research developed tools for extracting text with complex qualities such as dotted text printed on curved reflective material, text containing touching characters, poor background contrast, white, curved, rotated, and/or differing fonts or character width between sets of images. Computer simulation shows that the tools herein successfully handle recognition whether the text on the water bottles was raised, indented, or flat and/or if the text is shiny and the bottle material had a matte finish or vice versa.

All of the tools created by this research were integrated into specialized systems with different characteristics such as recognition systems for Ozarka® and Dasani® water bottles, concrete slabs, and license plates. In addition, the system's ability to read Arabic characters on the license plates illustrates the system's universality. These systems demonstrated successful recognition with an accuracy rate of 90-93%.