Line Detection in Run-Length Encoded Document Images using Monotonically Increasing Graph Model
Raveendra N Amarnath, Panduranga Naidu Nagabhushan, Mohammed Javed · 2018
Automatic detection of lines in images is vital, specifically in the field of document image analysis, because this facilitate the operations like segmentation, optical character recognition, and content based image retrieval. In the existing literature, line detection techniques have been demonstrated in the uncompressed version of the images, and also these techniques fail to work with compressed images. In today's digital era, as images are archived and transmitted in the compressed form, there is a necessity to develop novel techniques that can work directly and efficiently (time and space) with the compressed images without decompressing and recompressing the images. This research article presents a method to automatically detect straight lines in the run-length compressed document images. The technique is based on finding connected components of foreground region (runs) in a 2-dimensional run-length represented document image. Here, considering a minimum run in a column as a seed-node or root-node and tracing its neighbors along the columns by employing monotonically increasing graph model would result in detecting the straight lines. The proposed technique is tested with the various compressed document images, and subsequently experimental results and efficacy of the method are reported.