Script Identification of Movie Titles from Posters
Lists
Only you see this
Abstract
In this work, a new problem of script identification in movie posters has been addressed. Movie posters contain an amalgamation of different types of objects like images of actors, sceneries, different graphic symbols, several texts having disparate fonts, colors, textures, etc. Such a complex set of components makes it a challenging task in the automatic identification of the script of the movie titles for further processing. Before identifying the script of the titles, localization of the texts is very much necessary. Using transfer learning and non-maximum suppression the text localization has been performed followed textural feature-based script identification among Bangla, Devanagari, and Roman. We experimented with these poster images from Tollywood, Bollywood, and Hollywood and obtained the highest accuracy of 90.65%.
BibTeX
@inproceedings{ghosh2020script,
author = {Mridul Ghosh and Himadri Mukherjee and Sayan Saha Roy and Sk Md Obaidullah and KC Santosh and Kaushik Roy},
title = {{Script Identification of Movie Titles from Posters}},
booktitle = {RTIP2R},
year = {2020},
doi = {10.1007/978-981-16-0507-9\_10},
}Loading related content...