Word matching and retrieval from images

As vast amount of digital image data is stored by the advanced libraries, there is a requirement for an efficient query word searching methodologies which can make them accessible according to user's requirement. For their accurate retrieval, it is essential to understand their contents. Present technologies for optical character recognition (OCR) and image document analysis do not handle such documents adequately because of the recognition errors. Due to the problems faced by traditional OCR during recognition, computer is unable to extract the textual characters properly after scanning them. In this paper, we propose an effective word extraction and matching scheme from image documents that achieves high performance, even in the presence of noise in the image, degradation and font-variants. Initially, each image in image-database is pre-processed. In the next step, find contour method is used to detect blobs which are further passed in tesseract engine. Tesseract segments the characters from the image and stores in character database. Each word in the database is used to index a given set of images. During retrieval, the query word presented to the system is matched with characters in the database and all images containing instances of the query word are retrieved and presented to the user. Using this approach, our system is able to properly handle images with different font styles, size and heavily touching characters. From the experimental results on the various image formats it is observed that the extraction of text from the images is mostly accurate and indexing of words based on the position is working perfectly.