Scene Text Component Segmentation Using Hierarchical Distance Slicing
Payel Sengupta, Ayatullah Faruk Mollah · SSRN Electronic Journal · 2019
Segmentation of scene text components to recognize the text contained therein is an active area in scene image analysis. In this paper, a new character segmentation method has been proposed for segmenting scene text candidate regions extracted from scene images. At first, original scene text components are converted into binary text components. Then, we apply distance transformation on these components and generate the distance matrix. After that hierarchical distance slices are generated and total number of hierarchical distance slices are divided into different quarters. After slice selection we apply histogram based vertical projection profile to segment the characters one by one. This method is implemented on a dataset of 260 camera captured text components. It has been observed that the proposed method is very effective for adjacent characters which are either closely placed or are overlapping as well as disconnected. Experimental results show an average character segmentation rate of 80.69% for the dataset of closely placed and disconnected characters.