realizationofseuratâ€™s pointillism vianon...
Post on 11-Sep-2018
Embed Size (px)
Visual Comput (2007)DOI 10.1007/s00371-007-0183-y O R I G I N A L A R T I C L E
Chuan-Kai YangHui-Lin Yang
Realization of Seurats pointillismvia non-photorealistic rendering
C.-K. Yang () H.-L. YangDepartment of Information ManagementNational Taiwan University of Scienceand Technology,No. 43, Section 4, Keelung Road, Taipei,10607, Taiwanckyang@cs.ntust.edu.tw,D9509102@mail.ntust.edu.tw
Abstract Photorealistic rendering isone of the oldest and most importantresearch areas in computer graphics.More recently, the concept of non-photorealistic rendering has beenproposed as an alternative withimportant advantages for numerousapplication areas. The popularityof non-photorealism can be mainlyattributed to its simplicity, whichcould potentially lead to a state ofaesthetics and succinctness. Realityoften presents too many details andtoo much complexity, thus offsettingthe observation of the essence ofobjects, and objects interaction withlights. Based on a similar belief,impressionism focuses primarily onconveying the interaction of lightand shadows without emphasizingthe fine details of a scene. In recentyears, there has been a trend towardssimulating impressionism withcomputers. Among the various stylesof impressionism, we are particularlyinterested in simulating the styleof pointillism, especially the stylepresented by Georges-Pierre Seurat,deemed the founder of pointillism.
The reason his style attracts us istwofold. First, the painting processof pointillism is extremely laborious,so unloading the main proportion ofthe manual painting task is mostlydesired. Second, though severalexisting general-purposed algorithmsmay be able to approximate pointil-lism with point-like strokes, somedelicate features frequently observedin Seurats paintings are still notsatisfactorily reflected. To achievesimulating Seurats painting style, wehave made careful observations ofall accessible Seurats paintings andextracted from them some importantfeatures, such as the relatively fewprimitive colors, color juxtaposition,point sizes, and, in particular, theeffects of complementary colors andhalos. These features have been moresuccessfully simulated and results arecomparable with not only Seuratsexisting paintings, but also withprevious attempted simulations.
Keywords NPR Pointillism Halos Complementary colors Dithering
Traditionally there is almost no doubt that computergraphics technologies aim at mimicking reality. Sophis-ticated mathematics and physics are involved during thepursuit of reproducing the reality through computers. Suchan ambition has been nearly fulfilled in recent years due
to the blossoming of techniques on global illumination,texture mapping and image-based rendering. Surprisingly,there was a different trend emerging in the last decade:how to perform rendering in non-photorealistic styles. Thesuccess of such a dramatically different philosophy is notaccidental, but could be attributed to at least the followingtwo reasons. First, just like what happened in art history,
C.-K. Yang, H.-L. Yang
both realism and abstractionism have their supporters, butif we regard the computer graphics technology only asone of the many means to create images, then photoreal-ism needs not to be the only goal. Second, nowadays itis often deemed an era of information-explosion; there-fore, modern data tend to become huge and more complex,or sometimes even beyond human comprehension, andthis situation may get even worse with the consistent ad-vances on the data-acquisition process. Non-photorealisticrendering, or NPR for short, may serve as an alternativemeans to convey information in a more simplified and suc-cinct way, and examples of such applications can be foundin [10, 13].
As one of the most famous non-photorealistic paint-ing styles, impressionism concerns mainly the interactionbetween light and shadows, or what the so-called impres-sion that a scene can bring to a person, without placingtoo much emphasis on irrelevant details. There has beena trend towards simulating impressionism with comput-ers, either by converting existing images or videos intoones with impressionistic styles, such as [9, 12, 14, 1921, 32] or by designing a graphical painting interfacewhere images of impressionism styles could be easilycomposed, such as [5, 17, 18, 22]. Among the previouslysimulated impressionism styles, pointillism catches mostof our attention, and among the painters in pointillism,Seurat attracts us for the following two reasons. First, itis well known that a pointillistic painting usually takesa very long time to finish, as it involves a very tediousprocess which is potentially suitable for computers. Sec-ond, though generic NPR techniques exist to approximatepointillism effects, such as [12, 14, 19, 20, 32], they areoften too general to faithfully reflect all the features pre-sented in Seurats works. For example, in all Seuratspaintings, the number of colors in his color palette is veryrestricted, and some even said that essentially he used onlyeleven colors in his works. Furthermore, we observed thata feature called halo, that frequently appears in his paint-ings, has not been successfully simulated so far. Moreover,Seurat is also characterized by his use of color juxtapo-sition and complementary colors. To accurately emulateSeurats painting style, we have made careful and detailedobservations on numerous paintings of his in order to ex-tract the common or important features that distinguish hiswork from others. With these features properly addressed,a given image, accompanied by a users choice of sev-eral parameters, could be turned into a painting-like imagewith a presumably Seurats style. To prove the success ofour efforts, simulated images are presented at the end ofthe paper, together with their comparisons with the resultsproduced by other existing techniques. One of Seuratsown paintings, after digitization, is also shown for beingthe ground truth for comparison.
The rest of the paper is organized as follows. Section 2reviews some existing works related to this study. Sec-tion 3 briefly describes what we observed from Seurats
paintings, i.e., the distinct features. Section 4 discusseshow we simulate Seurats painting styles step by step. Sec-tion 5 presents our simulation results, which are comparedagainst previous simulations and one of Seurats paintings.Section 6 concludes this paper and hints for potential fu-ture directions.
2 Related work
The name of non-photorealistic rendering was firstbrought to the computer graphics community by Winken-back et al. in 1994 . Since then, a huge number of papershave been generated. As our main goal is to convert existingimages into pointillistic ones, we will only pay attention tothose similar studies. Readers who are interested in learningmore information regarding NPR could refer to .
Perhaps it would be better that we first distinguish be-tween stippling  and pointillism. In stippling, the focusis mainly on the placement of points, where every pointusually bears a similar outlook. To simulate an area witha darker shade, denser points will usually be disposed onthat area; on the contrary, a brighter area normally corres-ponds to a lesser number of points. In pointillism, pointscould vary not only in sizes, shapes, but also in colors. Wewill explain how we achieve pointillism in great detailsin Sect. 4.
Probably being the first work to simulate pointillisticeffects, Hertzmann proposed to mimic the hand-paintedfeatures by curved brush strokes, and in particular, he at-tempted some pointillistic effects by placing dense circleswith perturbed hue and saturation values . A lay-ered structure is employed to process an image layerby layer, with progressively smaller brushes. While eachlayer is a blurred version of the original, smaller brushesare only applied to where the original image differs fromthe blurred one. Numerous results, including source im-ages, are available, and will be compared with ours.Litwinowicz also proposed to simulate impressionism byvarying the attributes of strokes . In addition, he alsogave discussions on more attributes of a point, such asclipping, anti-aliasing, texturing, orientation, and so on,so that users could have more control on the simulationresults. Temporal coherence was also mentioned for deal-ing with videos. Similarly, Hays et al. also presented anelaborated scheme that could control various attributesof strokes and therefore lends itself to stylized images orvideos . Simulation results, including the originals,are also available for comparisons. Zhu et al. introduceda unified framework that could facilitate the control ofstroke attributes and the inclusion of other features to bedeveloped in the future . Hertzmann proposed a ratherdifferent approach for simulating a stylized painting byperforming image analogies ; however, we believethat his method cannot produce better results than that bya more domain-specific approach.
Realization of Seurats pointillism via non-photorealistic rendering
Most of the aforementioned approaches concentrateon simulating the strokes, while leaving color variationto a random perturbation. The work of Luong et al. adopted a different approach, where they pick colors in anisoluminant manner, so that the results match well withthe human perception system. Similarly, Jing et al. divided an input image into tiles, and within which col-ors are also perturbed in an isoluminant way. The workby Hausner  is probably the most similar work to oursso far. However, the half-toning patterns that he generatedby an error diffusion method seem too regular to create anartistic feeling. There is also not much variation in termsof point sizes and shapes, not to mention the lack of otherimportant features, such as halos tha