Panel Tracking for the Extraction and the Classification of Speech Balloons

Searching for texts inside a full comic strip may be exhaustive, and can be simplified by restricting the scope of the search to single panels, and better yet to within individual speech balloon. In this paper, a novel approach is devised where a tracking algorithm is employed for panel extraction, and speech balloons are identified using ‘Roberts’ edge detection operator as well as a classifier to find the number of balloons within every panel using a nonexhaustive projection method. Two main objectives in the field of comic strip understanding are achieved through our panel tracking for the extraction and classification of speech balloons (PaTEC). PaTEC may be incorporated as a precursor to text extraction and recognition reducing the computational time and effort of searching the whole image to the speech balloon area itself. PaTEC accuracy for panel extraction is 88.78% while balloon classification accuracy is 81.49% on a homegrown comic database.