End to end Bangla handwritten and scene text detection using convolutional neural network

This thesis report is submitted in partial fulfilment of the requirements for the degree of Bachelor of Science in Computer Science and Engineering, 2017.Cataloged from PDF version of thesis report.Includes bibliographical references (pages 27-28).Handwritten text detection from a natural image has a large set of difficulties. A systematic approach that can automatically recognise text from handwriting, printed books, road signs and also classifies text and nontext blocks from natural image has many significant applications. For instance, visual assistance for visually impaired people, image understanding, classification of text in image, implementing autonomous navigation system. Recent development of deep learning approach has strong capabilities to extract high level feature from a kernel(patch) of an Image. In this thesis we will demonstrate an alternate approach that integrates a multilayer convolutional neural network (CNN) with supervised feature learning .This approach allows a higher recall rate for the text in an image and thus increases the overall performances of the system. And we have used these methodologies to create a learning model using synthetic and real-world data that is capable to process bangla and english handwritten and scene text in natural image.Somania Nur MahalB M AbirFahim BakhtiarB. Computer Science and Engineerin