An Approach for Non-photorealistic Rendering That Is Appealing to Human Viewers

Non-photo realistic rendering is common to both Computer Graphics and Digital Image Processing, and encompasses a broad class of techniques for creating artistic images from scenes, geometric models or photographic images. In this context, this paper presents a framework designed to generate cartoon-like abstracted images, which combines techniques for region smoothing and segmentation, edge detection, adjustment of brightness/saturation and color harmonization. A subjective evaluation involving the output of related work was performed. Additionally, an experimental study involving human evaluators was conducted, in which the proposed approach was compared with five existing tools for image cartoonization. Statistical analysis performed on experimental data from one-way (1-F) ANOVA and Tukey-Kramer tests revealed significant differences in preference between results created by the proposed approach and those created by the competing approaches.