CPD: Faster RCNN-based DragonBall Comic Panel Detection
Rishabh Sharma, Vinay Kukreja
Abstract
A vital technique for addressing several necessities of mobile comic reading, such as comic content adaption, is comic panel extraction, or the breakdown of a comic page image into panels. The majority of current techniques are dependent on manually created, which limits their capacity to deal with erratic comic panel layouts. To address these limitations, the proposed work developed a comic panel detection (CPD) system based on a faster region-based convolutional neural network (RCNN) with non-maximum suppression (NMS) module on the comic image dataset of 6000 images prepared using various pre-processing techniques. The complete implementation technique has resulted in the detection accuracy of 84.19% for intersection of union-based results. Numerous experimental findings show that the suggested strategy significantly exceeds the case of the DragonBall dataset in terms of effectiveness and efficiency as determined by the F1 score and page accuracy. Additionally, the proposed study will benefit readers and researchers in the indicated domain by advancing technological transfer, community improvement, quality of life, and technological progress.