Text segmentation in color images using tensor voting

In natural scene, text elements are corrupted by many types of noise, such as streaks, highlights, or cracks. These effects make the clean and automatic segmentation very difficult and can reduce the accuracy of further analysis such as optical character recognition. We propose a method to drastically improve segmentation using tensor voting as the main filtering step. We first decompose an image into chromatic and achromatic regions. We then identify text layers using tensor voting, and remove noise using adaptive median filter iteratively. Finally, density estimation for center modes detection and K-means clustering algorithm is performed later for segmentation of values according to hue or intensity component in the improved image. Excellent results are achieved in experiments on real images. Ó 2006 Elsevier B.V. All rights reserved.

• Type: Journal
• Year: 2007
• Where: IVC
• Authors: Jaeguyn Lim, Jonghyun Park, Gérard G. Medioni
• Keywords: IVC 2007, Main filtering Step, Optical Character Recognition, Tensor Voting