Script independent text segmentation of document images using graph network based shortest path scheme

• Published in: International journal of information technology (Singapore. Online) Vol. 15; no. 4; pp. 2247 - 2261
• Main Authors: Sahare, Parul, Tembhurne, Jitendra V., Parate, Mayur R., Diwan, Tausif, Dhok, Sanjay B.
• Format: Journal Article
• Language: English
• Published: Singapore Springer Nature Singapore 01.04.2023; Springer Nature B.V
• Subjects: Algorithms; Artificial Intelligence; Computer Imaging; Computer Science; Dijkstra's algorithm; Documents; Handwriting recognition; Historic documents; Image processing; Image Processing and Computer Vision; Image segmentation; Machine Learning; Noise; Original Research; Pattern Recognition and Graphics; Shortest path planning; Software Engineering; Vision; Wavelet transforms; Words (language); Text segmentation; Dijkstra’s algorithm; Document image analysis; Script independent; Text processing
• ISSN: 2511-2104; 2511-2112
• DOI: 10.1007/s41870-023-01230-w

Document image processing is one of the growing research fields in the digital world for applications like data base indexing, text recognition, signature verification, web-searching engines, etc. Segmenting intermixed texts (handwritten and machine-printed) from documents is a difficult task. In this paper, script independent text-line and word segmentation techniques are proposed. For text-line segmentation, Dijkstra’s algorithm is employed, whereas for segmenting words, wavelet transform is used. Text-line segmentation is modeled as a general image segmentation task. Dijkstra’s algorithm is a shortest path planning method, which is utilized for boundary growing process. This forms potential text-line boundary regions. For word segmentation, energy map is calculated first using wavelet transform and further, Gaussian filter is used for text-blocks creation. Proposed techniques are evaluated on different databases contain documents of different scripts. Benchmarking analysis is performed with other approaches where highest segmentation accuracies of 97.6% and 98.1% are obtained by text-line and word segmentation techniques, respectively.