Paper Details: Downloads: 1256
Serial Number: P1151614489
Title: Text Skew Detection and Correction in Printed Text Images Relying on 2D Haar Wavelets
Authors: Tanwir Zaman and Vladimir Kulyukin and Adele Cutler
Abstract: A text skew detection algorithm is presented for printed text images. The algorithm applies the 2D Haar Wavelet Transform to an input image to compute the horizontal, vertical, and diagonal change matrices. The matrices are binarized and combined into a single matrix of intensity changes. The convex hull algorithm is applied to find a minimum area rectangle bounding the points in the matrix of intensity changes. The text skew is computed as the rotation angle of the bounding rectangle relative to the absolute north at 90 degrees. No constraints are placed on the magnitude of the text skew. The algorithm’s performance is compared with the performance of five text skew detection algorithms on 1001 U.S. nutrition label images and 2200 single- and multi-column document images in multiple languages. The experiments indicate that the proposed algorithm detects text skew angles in real time with an accuracy as high or higher than the accuracy of the other five algorithms. To ensure the reproducibility of the results reported in this article, the JAVA source code of the algorithm is made publicly available.
Keywords: computer vision, text skew detection, OCR, 2D Haar wavelet transform, mobile nutrition management
Journal/Conference: International Journal of Graphics, Vision and Image Processing
Volume: 16
Issue: 1
Submission Date: 3/29/2016 12:00:00 AM
Review Date: 4/11/2016 12:00:00 AM
Publishing Date: 5/31/2016 12:00:00 AM
Article Downloads: 1256
Download:

Facebook