Community tagging for mobile media

Abstract

The main objective of this project is to develop an automatic image annotation system as manual tagging of images is a cumbersome process. This report propose to annotate images based on text information available in the image since text are useful for describing the content of an image and is a powerful source of high-level semantics. The most direct way of extracting text from an image is to use a commercial OCR. However OCR is found to perform well only on simple background images where the contrast of background to text is high. The OCR is unable to handle images of complicated background. As such, preprocessing of images is needed prior to feeding it to OCR for text recognition. Such preprocessing includes text segmentation and binarization. Text segmentation is used to segment the text from the complex background and text binarization is used to enhance the contrast of background to text for optimal OCR performance. This report discuss the various approach to text segmentation and text binarization and concludes that text segmentation using edge and texture analysis and text binarization using joint entropy yields better performance. Finally, the text recognition output from the OCR will be further processed by a keyword extraction algorithm to extract suitable keywords for image annotation.