Character segmentation on traffic panels using visual appearance from roadside imagery
Sarwar Shah Khan, Somying Thainimit, Itsuo Kumazawa, Sanparith Marukatat, Faisal Ghaffar · 2017
Segmentation of text on traffic panels has shown great importance. Detection and recognition of traffic panels text is still a challenge in computer vision due to its different types and hug asymmetric information. This paper proposed a method to detect roadside traffic panels and extract the information on them. In the first stage, blue, white and green color segmentation and structure classification is done for the purpose of detecting traffic panels. Detected region is then extracted from source image. Based on traffic panels background color mask is used to remove non text candidates. Maximally stable extremal regions (MSER) identify the stable characters regions. Finally, MSERs from every level are combined to form the ultimate segmentation. The proposed method is evaluated on Thai traffic panels and experiments show that it performs well for both English and Thai character.