B05: Image Abstraction with Cartoonlike Shade Representation

Lee,G.R. , Lee,H.C. , Lee,T.M. , Yoon,G.H.

Luminance quantization, which maps the luminance values of an image to discrete levels, is widely used for image abstraction and the expression of a cartoonlike effect. Existing luminance quantization techniques use each pixel's luminance value separately, leading to a noisy image. Additionally, they do not take the shape of the imaged object into consideration. Thus, they suffer limitations in terms of cartoonlike shade representation.
We propose a new luminance quantization algorithm that takes into account the shape of the image. We extract the silhouette from the image, compute edge-distance values, and incorporate this information into the process of luminance quantization. We integrate the luminance values of neighboring pixels using an anisotropic filter, using gradient information for this filtering. As a result, our quantized image is superior to that given by existing techniques.