Investigation of robust color model for edge detection on text extraction from scenery images

This paper discusses a robust method to detect the edges of text in natural scenes, which is against the effects caused by shadow or highlight. Many color models are tested to find the robust ones under arbitrary image conditions. Moreover, some new color models, which are invariant to shadow or highlight, are proposed. By effectively utilizing color information, a simple edge threshold decision scheme is presented. After that a combined scheme is introduced by virtue of the invariant characteristics of each color model in it, which could get almost intact edge information for text region. It shows good performance for the text extraction against shadow and highlight. Based on above, a simple and robust method for edge detection is achieved, which is useful for text extraction and recognition tasks.