Box Flame Detection and Image Normalization in Comic Images

In recent years, digital comic images are popular on the Internet. In general, they are scanned from comic books, but most of them are not normalized about their size, skew and margin. The normalization of comic images in preprocessing step has many advantages in automatic comic image analysis such as content-based comic image retrieval systems. The normalization in comic image can be carried out efficiently by using box frames in comic images. In this paper, we propose a comic image normalizing method using the box frames which are detected by line segments extracted by the gradient Radon transform. We applied the Radon transform using image gradients to extract line segments which are candidate line segments of the box frame sides. The final box frames are detected by cutting out regions which do not have comic image contents. By using region homogeneity, we decide whether the region has contents in comic images or not. Finally, the comic images are normalized by the homography transformation using the box frame extracted. Experimental results show that our method effectively detects the box frames and normalizes comic images.