On the effects of pre- and post-processing in video cartoonization with bilateral filters

Abstract

In recent years, advances in image-based artistic rendering have grown steadily with the additional leverage of image and video processing techniques. Video cartoonization or stylization is the process of artificially incorporating cartoon-like effects to photorealistic input videos. This paper investigates the effects of integrating relevant pre- and post-processing tasks to significantly improve the quality of cartoonized videos processed with bilateral filters (BF). Our video cartoonization framework work extends the original Winnemöller's real-time video abstraction framework, which applies the edge-preserving BF with additional use of edge maps, luminance quantization and frame temporal coherency. In our work, we propose a contrast enhancement option by intensity stretching and Laplacian filtering to finetune the contrast levels of the pre-BF frames. For the post-BF recombined frames, an unsharp masking procedure is proposed to accentuate feature details in the final output video. Results from extensive experiments conducted by qualitative user evaluation underline the essentiality of pre- and post-processing tasks for improved video cartoonization.