TEXT DETECTION FROM IMAGES

Abstract

Images and videos contain information that is useful for indexing and annotation of multimedia data. Steps involved in extracting this information are detection and localization of possible text candidates, eliminating false positives, enhancing the remaining text candidates and then recognizing them using an Optical Character Recognition. But, the text that is contained in each image or video frame can have different color, size, illumination, font, orientation and texture. All these attributes make text detection and recognition a challenging task. Some of the known text detection approaches have been discussed in the literature survey and out of these techniques, Maximally Stable Extremal Region algorithm has been found to be the best region detectors. It has been coupled with Edge detection methods to improve the accuracy of results.