Videos

Scene-text video demonstrates the importance and requirement of text localization in scene images. A sample scene image is fed to OCR engine (Here, Tesseract-OCR is used for the demonstration), the text, which is present in the image, is not recognized properly. If text localization is performed before passing it to OCR, then the text can be recognized. In the video, text is manually located and NESP algorithm is used to segment manually located text. The segmented text is passed to Tesseract-OCR for recognition.

A sample application of scene-text analysis is capture a movie poster on roadside or in cinema halls. Recognize the movie name and get additional information on movie like release date, screening places and many more. Man of steel movie poster is used to demonstrate the application.

Text segmentation and localization by OTCYMIST algorithm. OTCYMIST algorithm won First place for performance of text segmentation task on Born-digital images conducted in ICDAR 2011 as Challenge 1. A demo of OTCYMIST algorithm is shown as video.

Word, manually cropped from scene or born-digital images are difficult to recognize. For recognition, we segment the word images and pass it to OCR. Since, the recognition of segmented words is easier for an OCR. Here, complex word images with degradations like texture, varying illumination and blur are recognized by segmenting them. Segmentation results are also shown in the video along with the recognized text.

Kannada words are also recognized from manually cropped scene images. Kannada OCR is built for the purpose of word recognition. Some characters are classified as other characters. More words had a edit distance of one from ground-truth Unicode word. Video include properly recognized word as well as one edit distance words.