Text Extraction and Localization From Captured Images

Abstract

Extraction of text contents from image is tedious task because of variance in the font size, style, Orientation, Alignment and heterogeneous nature of text. The contents of text information in the scene images hold valuable data. The framework uses 2-d Wavelet transform using HAAR is applied to the grayscale image followed by edge detection for each sub-band filtering. Then region clustering technique is applied using centroids to each region. Further bounding box is fit to each region thus identifying the text components. Proposed framework is removing non-text content and separated text-content, then each text contents are converted into editable text using OCR engine. Here, we use Teserract recognition engine.