Unsupervised Text Extraction from G-Maps

TL;DR

This paper introduces an unsupervised method for extracting text from Google Maps and GIS images, utilizing image segmentation, edge detection, and connected component analysis to achieve high accuracy without prior training.

Contribution

It presents a novel unsupervised approach combining Fuzzy CMeans, Prewitt edge detection, and connected component analysis for effective text extraction from maps.

Findings

Achieves 98.5% accuracy on experimental datasets

No prior training or knowledge required

Effective for GIS and Google Maps images

Abstract

This paper represents an text extraction method from Google maps, GIS maps/images. Due to an unsupervised approach there is no requirement of any prior knowledge or training set about the textual and non-textual parts. Fuzzy CMeans clustering technique is used for image segmentation and Prewitt method is used to detect the edges. Connected component analysis and gridding technique enhance the correctness of the results. The proposed method reaches 98.5% accuracy level on the basis of experimental data sets.