Localization and Extraction of Qur’an Verses Using Computer Vision

Abstract

Localizing Quranic verses, by detecting the verse bounding boxes, with respect to Quran page images is crucial for UI applications. These applications rely on the user interacting with the verse to view the translation, share the verse, listen to its audio, etc. Moreover, the automatic detection of the verse bounding boxes enables additional image processing and analysis of the Quran pages at the verse level. For these use cases, we need to map the user's click within the image boundary and know which verse is selected. In this paper, we propose a computer vision approach using a Faster RCNN neural network to analyze Quran page images and automatically localize the boundary of every verse with respect to the page. This information can, later on, be fed into various UI applications that allow the user to interact with Quran verses. We train our model and run several experiments on the following narrations: Hafs, Douri, Shubah, Qalon, and Warsh. Our results show 100% accurate detection of all verse boundaries for these narration