Effective Mode of Learning Cartonization: White-box Cartoon Representations

Abstract

Cartoon, a renowned art form has different forms of application in various scenarios. In essence, different cartoon styles and use cases, demand prior knowledge to develop a usable algorithm or they have certain assumptions that needs to developed which cater to a particular task. The present paper’s objective is to determine an impactful approach towards image cartoonization, by developing an efficient and effective mode of such learning. The author proposes to identify three white-box representations from images individually, namely, the surface representation that covers a smooth surface of cartoon images, the structure representation that observes the sparse color-blocks and the texture representation which observes high- frequency texture, contours, details in cartoon images. The objectives of the proposed method are separately based on each extracted representations, that applies a Generative Adversarial Network (GAN) framework to understand the extracted representations and cartoonize images, which further enables the framework to be under control and adjustable at the convenience of the author. The present approach aspires to benefit the requirements of an artist that deals in several cases and has various style requirements. The author has conducted analytical and doctrinal studies, coupled with bibliometric research, that covers both quantitative analysis and qualitative comparisons, accompanied with the empirical study that have been present in several research papers, related to the subject have been thoroughly analyzed by the researcher to add something to existing knowledge on White box Cartoon Representations. Lastly, it is observed that the ablation study carried out shall showcase an influence of each component in the framework suggested by the author.