Line-wise text identification in comic books: A support vector machine-based approach
Document Type
Conference Article
Publication Title
Proceedings of the International Joint Conference on Neural Networks
Abstract
This paper presents a study of line-wise text identification in comic books. Due to the unavailability of a single OCR system which can handle comic text of multiple scripts, the comic text identification based on script becomes an essential step for choosing the appropriate OCR. In this investigation, a new attempt has been made to explore a comic text identification technique of speech balloon to feed the identified text into the appropriate OCR. Latin and Bengali comic text lines have been considered for identification in this study. Two different local features, namely, Scale Invariant Feature Transform (SIFT) and Multi-scale Block Local Binary Pattern (MBLBP) were considered in Spatial Pyramid Matching (SPM) domain in the current study. The support vector machine (SVM)-based classification technique has been considered for line-wise text identification. To evaluate the identification system, text datasets of Latin and Bengali comics have been newly prepared from Latin comic e-books and Bengali comic books respectively. The Latin comic e-books are collected from internet on website dedicated to Comics. To conduct the experiment, samples of 1938 text lines from each comic text dataset have been used. A publicly available eBDtheque comic text database has also been considered for performance comparison of the proposed method. 1938 number of text line images from eBDtheque comic text database has also taken into account in this approach. The highest identification accuracies of 98.30% and 98.29% on an average are achieved in the experiment when Bengali and eBDtheque comic text dataset are considered.
First Page
3995
Last Page
4000
DOI
10.1109/IJCNN.2016.7727719
Publication Date
10-31-2016
Recommended Citation
Pal, Srikanta; Burie, J. Christophe; Pal, Umapada; and Ogier, Jean Marc, "Line-wise text identification in comic books: A support vector machine-based approach" (2016). Conference Articles. 739.
https://digitalcommons.isical.ac.in/conf-articles/739
