A tree conditional random field model for panel detection in comic images

• A novel context modeling method is used to detect panels within comic images.
• Connected components, edge and line segments are extracted from comic images.
• We extract geometric and contextual features for each of the three kinds of objects.
• A tree CRF is used to model the interactions among these objects.
• The detected panels are presented as convex polygons to keep the content integrity.

The goal of panel detection is to decompose the comic image into several panels (or frames), which is the fundamental step to produce digital comic books that are suitable for reading on mobile devices. The existing methods are limited in presenting the extracted panels as squares or rectangles and solely use one type of visual patterns, which are not generic in terms of handling comic images with multiple styles or complex layouts. To overcome the shortcomings of the existing approaches, we propose a novel method to detect panels within comic images. The method incorporates three types of visual patterns extracted from the comic image at different levels and a tree conditional random field framework is used to label each visual pattern by modeling its contextual dependencies. The final panel detection results are obtained by the visual pattern labels and a post-processing stage. Notably, the detected panels are presented as convex polygons in order to keep their content integrity. Experimental results demonstrate that the proposed method achieves better performance than the existing ones.