Multispectral Image Matting of Ancient Chinese Paintings

Abstract

Digital matting, the process of extracting a foreground object from an image, is an important task in image and video editing. Applying matting techniques to Chinese painting image processing can create novel composites or facilitate other editing tasks. However, Chinese paintings are painted on xuan-paper or silk, the semi-transparent strokes resulted from the diffusion and penetration of ink and pigments make it difficult to extract the foreground from the paintings only based on three-band image. In this paper, we demonstrate a new multispectral image matting technique for Chinese painting image editing. We derive a similarity function from Kubelka-Munk turbid media theory, and this allows us to find the optimal alpha matte. By adopting multispectral matting method, semitransparent foreground stroke can be extracted from the overlay of background strokes. Experimental results show the approach acceptable and promising.