Watercolour Rendering of Portraits

Abstract

Abstract. Applying non-photorealistic rendering techniques to stylise portraits needs to be done with care, as facial artifacts are particularly disagreeable. This paper describes a technique for watercolour rendering that uses a facial model to preserve distinctive facial characteristics and reduce unpleasing distortions of the face, while maintaining abstraction and stylisation of the overall image, employing stylistic elements of watercolour such as edge darkening, wobbling, glazing and diffusion.

Benchmarking non-photorealistic rendering of portraits

“…To date, these benchmark datasets have been used in a variety of ways: to include some stylisation results from examples taken from the benchmark [16], [17], [18], [19]; to provide appropriate test data as part of the optimisation of preset parameters for post-processing filters in BeCasso, an interactive mobile iOS app for image stylisation [20]; and to provide a competitive and common set of test images for a research course on image processing for mobile applications [21].…”

“…Table 8 shows the Pearson and Kendall correlation coefficients; the values confirm that general-purpose filtering approaches such as XDoG [42] and oil painting [48] are not affected by the increasing complexity across the benchmark levels. Although they are face-specific, watercolour [18] and engraving are also fairly robust since their renderings are not highly dependent on the face model, and their results are reasonable despite inaccurate face detection. The techniques with highest correlation to the levels are neural style transfer [40], which has a tendency to create more spurious facial features (e.g.…”

“…Rosin and Lai [18] use a filter based approach to generate a watercolour stylisation. In order to achieve the multiple characteristics of watercolour -namely brightening, abstraction, edge darkening, wobbling, granulation, glazing, pigment and paper variations -they employ various steps such as smoothing, morphological opening and closing, contrastlimited local histogram equalisation, edge detection, overlay blend, local geometric distortion, superpixel segmentation, and level of detail masks controlled by face detection and saliency masks.…”

“…We conducted Experiment 1 described in section 3.6 and applied it to 11 NPR algorithms which cover a wide range of styles and methods: neural style transfer [40], XDoG [42], oil painting [48], pebble mosaic [49], artistic sketch method [41], APDrawingGAN [43], puppet style [44] engraving [46], hedcut [47], Julian Opie style [44], watercolour [18]. The 11 NPR algorithms are run on the full 60-image benchmark and so the first user study to collect the four face characteristics contained 660 stylised photos.…”

“…Images from the NPRportrait1.0 benchmark stylised as pebble mosaics: Doyle et al[49] Images from the NPRportrait1.0 benchmark stylised as watercolours: Rosin and Lai[18] …”