Handwritten Text Segmentation Approach in Historical Arabic Documents

The sharp increase in the number of historical documents available as images in national libraries only increases their storage capacity. The need for access to content of this cultural heritage is also increasing. The methods of indexing and searching in image content are still very limited. Text segmentation of digital historical documents is an important step in recognizing content as images. In this paper, we present an original method of segmentation of lines and pseudo-words in historical documents based on Gaussian filters. Our method consists in detecting elliptical blobs in scales formed by Gaussian filters. Experimental tests are performed on hundreds of pages of historical documents. The experimental results showed with this method are excellent in front of other methods of text segmentation in manuscript images.