Identifying the Presence of Graphical Texts in Scene Images using CNN

Scene images contain graphical texts that usually carry dynamic information. Therefore, knowing the presence of graphical texts in images can help advance further processes, such as graphical text mining and recognition. The goal of the study is to distinguish scene images with graphical text from those without graphical text using the convolution neural network (CNN) based classification framework. For validation, four different publicly available image datasets: ICDAR2003, MSRA-TD500, SVT and 15-Scene were used and the results were encouraging.