Logo Detection in Arabic Documents Using Multi Smearing Method and Decision Tree

Abstract

The detection of logo techniques play significant role for document image analysis and retrieval. In this paper, an effective logo detection method in Arabic document images has been proposed. In the proposed technique different logos can be detected based on extracting features that will distinguish logo from other non-logo parts of document like text, graph, table, and also stamp. This model is divided into three main stages. The first stage is smearing stage, where the document image has been smeared in multi directions to segment image to different blocks represent foreground objects of document. The second stage is to extract appropriate and significant features from these blocks by bounding blocks into rectangles. The third stage is performing decision tree that consist of a number of rules that will be applied to block features to correctly classify logo from non-logo objects. The proposed technique overcome many problem of logo detection like logos that contains separated parts, logos with text, and logo with noise. This technique has been tested and evaluated on dataset containing variety of Arabic document images of different colors, shapes and resolutions. Experimental results exhibit its performance in detecting logos with 96% for accuracy and 98% for precision.