Text segmentation from scanned land map images using radon transform based projection profile

Paper land maps are useful documents for collecting geographical information. Land map images provide information like place names, country or states borders etc. A land map image can be considered as a complex document image as the texts may have the myriad background consists of various intensity values and different orientations. Text orientations, fonts and sizes may not unique for all place names within same map images. Text segmentation from such images is not an easy task. This paper presents an approach for text segmentation from scanned land map images using Radon transform based projection profile. Using this projection profile, localization of the text positions are done. Then Euler number of each component is computed for eliminating the non-text components. The proposed approach is tested on a set of collected map images (containing texts in Indian regional scripts; like Bangla) and it gives an encouraging result.