Semantic Document Image Classification Based on Valuable Text Pattern

Subject Areas : Renewable energy

Hossein Pourghassem ¹ , Mohammad sadegh Helforoush ² , Sabalan Daneshvar ³

1 - Assistant Professor/Islamic Azad University - Najafabad Branch
2 - Assistant Professor/ Shiraz University of Technology
3 - Assistant Professor/Sahand University of Technology – Tabriz

Received: 2009-05-14 Accepted : 2010-08-11 Published : 2011-02-01

Keywords: semantic classification, document and non-document images, information valuable,

Abstract :

Knowledge extraction from detected document image is a complex problem in the field of information technology. This problem becomes more intricate when we know, a negligible percentage of the detected document images are valuable. In this paper, a segmentation-based classification algorithm is used to analysis the document image. In this algorithm, using a two-stage segmentation approach, regions of the image are detected, and then classified to document and non-document (pure region) regions in the hierarchical classification. In this paper, a novel valuable definition is proposed to classify document image in to valuable or invaluable categories. The proposed algorithm is evaluated on a database consisting of the document and non-document image that provide from Internet. Experimental results show the efficiency of the proposed algorithm in the semantic document image classification. The proposed algorithm provides accuracy rate of 98.8% for valuable and invaluable document image classification problem.