An efficient hierarchical method for text region extraction in degraded document images

This paper presents a clustering based method to extract text regions from degraded document images. In this method the gray scale image is decomposed into four sub-bands using discrete wavelet transform. For each pixel, the corresponding components of 3 detail sub-bands are considered as feature vector. Potential text regions are extracted by k-means clustering algorithm. We propose several heuristic constrains by which candidate text regions are refined to eliminate non-text regions. Evaluation over a set of degraded documents captured with camera, our method shows satisfactory results.

Keywords:

text extraction , degraded document , clustering , discrete wavelet transform , document segmentation