APPROACH TO TEXT EXTRACTION FROM IMAGE
Disha Bhat, Dornala Charitha, M K Dimple, R Amruthashree, G. H. Shruthi · International Journal of Advanced Research in Computer Science · 2018
The multimedia resources in a database and on the web are increasing. The multimedia resources can be images and videos. It has become a very difficult task to develop the effective methods to manage as well as to retrieve these resources by their content. The Text which is an important object which carries high-level semantic information which is useful for this task. The current technologyis optical character recognition (OCR) is used to convert machine generated text which is printed against clean background to computer readable form (ASCII). But, text isoften printed against shaded or textured backgrounds or is embedded in images.Examples include maps, photographs, advertisements, videos etc. Current document segmentation and recognition technologies cannot handle these situations well. Our system takes advantages of the distinctive characteristics of text that make it stand out from other image material that is, text possesses certain frequency and orientation information. We will first clean the image by changing the contrast and gradient of the image. Now the objects in the images are identified and numbered. Further in the text recognition process, these numbered objects are segregated into text and non-text. Later the recognised text is reconstructed to form a meaningful text present in the image. Also we are focusing on extracting the text such that certain portion of the images such as logos etc is retained. This is done by calculating the pixels of the required portion of the image to be retained and then training the system in such a way that it extracts all the text except the portion of the image to be retained.