Monograph. Computer Vision. (published jointly with Novática*) Guest Editors: Dídac López-Viñas, Marc Bigas-Bachs, Víktu Pons-Colomer,

Transcription

1 CEPIS UPGRADE is the European Journal for the Informatics Professional, published bimonthly at < Publisher CEPIS UPGRADE is published by CEPIS (Council of European Professional Informatics Societies, < in cooperation with the Spanish CEPIS society ATI (Asociación de Técnicos de Informática, < and its journal Novática Vol. XI, issue No. 6, December 2010 CEPIS UPGRADE monographs are published jointly with Novática, that published them in Spanish (full version printed; summary, abstracts and some articles online CEPIS UPGRADE was created in October 2000 by CEPIS and was first published by Novática and INFORMATIK/INFORMATIQUE, bimonthly journal of SVI/FSI (Swiss Federation of Professional Informatics Societies) CEPIS UPGRADE is the anchor point for UPENET (UPGRADE European NETwork), the network of CEPIS member societies publications, that currently includes the following ones: inforewiew, magazine from the Serbian CEPIS society JISA Informatica, journal from the Slovenian CEPIS society SDI Informatik-Spektrum, journal published by Springer Verlag on behalf of the CEPIS societies GI, Germany, and SI, Switzerland ITNOW, magazine published by Oxford University Press on behalf of the British CEPIS society BCS Mondo Digitale, digital journal from the Italian CEPIS society AICA Novática, journal from the Spanish CEPIS society ATI OCG Journal, journal from the Austrian CEPIS society OCG Pliroforiki, journal from the Cyprus CEPIS society CCS Tölvumál, journal from the Icelandic CEPIS society ISIP Editorial TeamEditorial Team Chief Editor: Llorenç Pagés-Casas Deputy Chief Editor: Rafael Fernández Calvo Associate Editor: Fiona Fanning Editorial Board Prof. Vasile Baltac, CEPIS President Prof. Wolffried Stucky, CEPIS Former President Hans A. Frederik, CEPIS Vice President Prof. Nello Scarabottolo, CEPIS Honorary Treasurer Fernando Piera Gómez and Llorenç Pagés-Casas, ATI (Spain) François Louis Nicolet, SI (Switzerland) Roberto Carniel, ALSI Tecnoteca (Italy) UPENET Advisory Board Dubravka Dukic (inforeview, Serbia) Matjaz Gams (Informatica, Slovenia) Hermann Engesser (Informatik-Spektrum, Germany and Switzerland) Brian Runciman (ITNOW, United Kingdom) Franco Filippazzi (Mondo Digitale, Italy) Llorenç Pagés-Casas (Novática, Spain) Veith Risak (OCG Journal, Austria) Panicos Masouras (Pliroforiki, Cyprus) Thorvardur Kári Ólafsson (Tölvumál, Iceland) Rafael Fernández Calvo (Coordination) English Language Editors: Mike Andersson, David Cash, Arthur Cook, Tracey Darch, Laura Davies, Nick Dunn, Rodney Fennemore, Hilary Green, Roger Harris, Jim Holder, Pat Moody. Cover page designed by Concha Arias-Pérez "The Circular Look" / ATI 2010 Layout Design: François Louis Nicolet Composition: Jorge Llácer-Gil de Ramales Editorial correspondence: Llorenç Pagés-Casas <pages@ati.es> Advertising correspondence: <info@cepis.org> Subscriptions If you wish to subscribe to CEPIS UPGRADE please send an to info@cepis.org with Subscribe to UPGRADE as the subject of the or follow the link Subscribe to UPGRADE at < Copyright Novática 2010 (for the monograph) CEPIS 2010 (for the sections Editorial, UPENET and CEPIS News) All rights reserved under otherwise stated. Abstracting is permitted with credit to the source. For copying, reprint, or republication permission, contact the Editorial Team The opinions expressed by the authors are their exclusive responsibility ISSN Monograph of next issue (February 2011) "Internet of Things" (The full schedule of CEPIS UPGRADE is available at our website) Monograph (published jointly with Novática*) Guest Editors: Dídac López-Viñas, Marc Bigas-Bachs, Víktu Pons-Colomer, and László Szirmay-Kalos 2 Presentation. Imaging Revolution Dídac López-Viñas, Marc Bigas-Bachs, Víktu Pons-Colomer, and László Szirmay-Kalos 5 GPU-based Ambient Occlusion and Indirect Illumination Balázs Tóth, Tamás Umenhoffer, László Szirmay-Kalos, and Mateu Sbert 15 Three-Dimensional Perception, Measuring Reality Joaquim Salvi 17 Technologies for 3D: Looking at the Future. Interview with Steve Schklair, Founder and CEO of 3ality Digital Jordi Alonso 20 Non-Photorealistic Rendering for Motion Picture Production Tamás Umenhoffer, László Szécsi, Milán Magdics, Gergely Klár, and László Szirmay-Kalos 28 On Creativity in Multimedia: "Serious Games" Oscar García-Pañella, Emiliano Labrador-Ruiz de la Hermosa, Anna Badia-Corrons, and Pau Moreno-Font 33 20,000 Photographs under the Sea Rafael García UPENET (UPGRADE European NETwork) 35 From Informatik Spektrum (GI, Germany, and SI, Switzerland) Society Consumer and Corporate Credit Ratings and the Subprime Crisis in the U.S. with Some Lessons for Germany Akos Rona-Tas and Stefanie Hiß CEPIS NEWS 49 Selected CEPIS News Fiona Fanning * This monograph will be also published in Spanish (full version printed; summary, abstracts, and some articles online) by Novática, journal of the Spanish CEPIS society ATI (Asociación de Técnicos de Informática) at <

2 Non-Photorealistic Rendering for Motion Picture Production Tamás Umenhoffer, László Szécsi, Milán Magdics, Gergely Klár, and László Szirmay-Kalos A critical problem of movie rendering is that while frames are calculated independently, the image sequence should not present flickering or inconsistency. A piece of curve must be consistent with the 3D geometry and move and deform with the object. On the other hand, it must also be consistent with the 2D image since in the traditional drawing process the artist draws in 2D. Hand drawn lines are roughly independent of the distance of the object from the viewer since rendering happens directly on the paper. Finally, manual drawing involves a great deal of randomness - the artist is not able to draw the same line twice. Being synchronized in both object and also in image spaces seems to be contradicting requirements. The rendering process should resolve this contradiction while presenting the natural randomness inherent in manual work. This paper presented an NPR (Non-Photorealistic Rendering) system and its application in the movie rendering pipeline. Our system implements three NPR effects: contouring, hatching, and painterly rendering. The algorithms that have been developed guarantee that these effects remain consistent with the animation and the deformation of the 3D geometry, avoid flickering, while still maintaining the look of hand drawn images. This project also proves that NPR effects can be made flexible enough to allow the artist to express his/her own ideas and produce images meeting his/her expectations. Keywords: 3D Rendering, Computer-generated Graphics, Photo-realism, Rendering, Shading. 1 Introduction Photo-realism has been the focus of rendering for decades. Photo-realistic rendering aims at creating images that are indistinguishable from real-world photographs, which is made possible by the precise simulation of physics laws (e.g. the Maxwell equations) during the rendering process. The level of accuracy of the representation of physics in the rendering code determines the level of realism of the result. In addition to the objective of photo-realism, computer graphics also tries to mimic artistic expression and illustration styles that do not create a photo-like impression. Such methods are usually vaguely classified as non photo-realistic rendering (NPR). While the fundamentals of photorealistic rendering in optics are well understood, NPR systems simulate artistic behavior that is not mathematically founded and often seems to be unpredictable. It means that the first step of NPR should be to gain an understanding of the artist, establishing a mathematical model describing his style, and finally solving this model with the computer. The Authors Tamás Umenhoffer received his Ph.D. from the Faculty of Electrical Engineering and Information Technology at the Budapest University of Technology and Economics, Hungary, in The topic of his dissertation was global illumination for sampled geometry on the Graphics Processing Unit (GPU). He is currently an assistant professor at the same university, and also works for Lichthof Productions on the development of Non- Photorealistic Rendering (NPR) in a feature film. <umitomi@gmail.com> László Szécsi is a professor adjoint of the Department of Control Engineering and Information Technology at the Budapest University of Technology and Economics, Hungary. He received his Ph.D. from this university in His research focuses on real-time global illumination and the exploitation of the GPU. <szecsi.laszlo@gmail.com> Milán Magdics is a Ph.D. student of the Department of Control Engineering and Information Technology at the Budapest University of Technology and Economics, Hungary. He is about to finish his Ph.D. programme and submit his research results on GPU based parallel Positron Emission Tomography reconstruction. In addition to General-Purpose computation on Graphics Processing Units (GPGPU) applications he has published several papers on procedural generation of the rendering on the shader processors. <gumi@inf.elte.hu> Gergely Klár is a Ph.D. student of the Eötvös Lóránd Science University, Hungary, where he works on natural phenomena simulation, GPU based solution of differential equations, and aesthetics modelling using special curves. <shinjin.cr@ gmail.com> László Szirmay-Kalos is head of the Department of Control Engineering and Information Technology at the Budapest University of Technology and Economics, Hungary. He received his Ph.D. in computer graphics from this university in 1992, and full professorship in His research areas are Monte- Carlo global illumination algorithms, tomography reconstruction, and the application of GPU in rendering and general purpose computations. He has more than two hundred publications in this field. He has spent long periods as a visiting professor at the Vienna University of Technology, Austria, University of Minnesota, USA, and Universitat de Girona, Spain. He is the fellow of Eurographics. <szirmay@iit.bme.hu> 20 CEPIS UPGRADE Vol. XI, No. 6, December 2010 Novática

3 result will be acceptable if our model is close to artistic behavior which cannot be precisely specified. During the history of NPR, many individual styles were simulated, and the developed algorithms were better or poorer in providing the illusion that a certain artist made this image by hand. In order to exploit the potential of NPR, these algorithms should be integrated in a complete system and, more importantly, their flexibility should be increased. Rendering with one particular style is not sufficient; the system must provide the artist with controls to create images that meet his own expectations. Motion picture production has a well established pipeline, modelers and animators define the 3D models, motions and lighting in a modeling system like Maya, a 3D rendering system (e.g. RenderMan) produces the sequence of layer images on a render farm, and finally the layers are composited in a 2D post-production framework. While the input of this process is a 3D world, the artist is traditionally used to working in 2D, on a piece of paper. Thus, while we wish to incorporate artistic styles into the production pipeline, the inherent dimensionality of particular techniques must be preserved. This requirement, and the possibility of interactive control, makes the post-production system the right place of the interactive definition of NPR elements. However, as the artist also takes into account 3D geometry and perception during his work, this information should also be made available and exploited during the NPR rendering. A critical problem of movie rendering is that while frames are calculated independently, the image sequence should not present flickering or inconsistency. A piece of curve must be consistent with the 3D geometry and move and deform with the object. On the other hand, it must also be consistent with the 2D image since in the traditional drawing process the artist draws in 2D. Hand drawn lines are roughly independent of the distance of the object from the viewer since rendering happens directly on the paper. Finally, manual drawing involves a great deal of randomness; the artist is not able to draw the same line twice. Being synchronized in both object and also in image spaces seems to be contradicting requirements. The rendering process should resolve this contradiction while presenting natural randomness inherent in manual work. This paper presents an NPR renderer that is integrated into a production pipeline and allows interactive control during the post-production phase. We consider three NPR techniques in the paper: Contours that highlight object silhouettes and lines that are important to express the surface geometry. Hatching that depicts shades and geometry by line primitives. The density of the lines corresponds to shades, the orientation to the surface geometry. Painterly rendering where distant or out of focus objects are blurred with simulating sparse brush strokes while close objects are finely presented. 2 Previous Work Painterly rendering covers the image by brush stokes to produce a painted look. Brush strokes may be fully random, sampled according to only the 2D image [1][2] or 3D geometry [3]. The size and the texture of the brushes can vary to simulate the work of an artist who only puts details in those objects that are put in the focus of the image. Line art illustration or pen-and-ink rendering uses textured line primitives to describe the scene. These line primitives can emphasize object boundaries, called silhouettes, and places where there is an abrupt change in surface geometry, e.g. in ridges or contours [4]. Lines can also express shades. Hatching is an artwork where the scene and its illumination conditions are depicted by line primitives [5]. The directions of the line primitives are roughly similar on an object surface around a point. If we do not wish to express information with the average direction, it can follow the isoparametric lines of the parametric surface or the gradient of the illumination. The best images can be obtained with following the principal curvature directions of the surface [6]. The principal curvature directions of a surface are those where the normal curvature of the surface has its maximum and minimum. In order to obtain these directions, we should find a local quadratic approximation of the surface, i.e. its second fundamental form, and compute the eigenvectors of the Weingarten matrix. If the surface is defined by a triangle mesh, then the quadratic approximation must be fit to a set of triangles close to the inspected point [7]. In order to make the lines consistent with the 3D objects, hatches can be generated into textures, which are then mapped onto the animated object. Unfortunately, this approach does not provide natural line-art, where each line has roughly similar width, which is determined by the pen of the artist and is independent of the depicted object s distance. From another point of view, the fact that line-art is naturally produced in image space implicitly assumes a level-of-detail mechanism, which renders objects with fewer lines if they are farther away and thus cover just a smaller portion of the image [8]. To guarantee consistency with the image, hatches can be placed directly in image space. However, the image space technique creates an annoying artifact, called the shower door effect when the camera is also animated. In this case, the user has the impression that hatch lines are not fixed to objects but are swimming on the image. Meier [3] proposed the application of particles attached to objects to provide the consistency with the animated 3D geometry. However, particle density is also affected by the image space distribution. Cornish et al. [9] controlled particle density using a mesh simplification algorithm. Unlike interactive systems [10], the movie production pipeline is a one way process. Passing information back from later stages to earlier ones is either impossible or complicated, and in any case expensive. Thus, the artist s input made in the post-production stage cannot directly affect the 3D rendering process. Instead, the rendering process should anticipate what the artist might need about the 3D scene and pass all information to the compositing phase. Depend- Novática CEPIS UPGRADE Vol. XI, No. 6, December

4 Figure 1: Comparison of the Conventional Rendering Pipeline and our NPR Pipeline where NPR Algorithms are integrated Into the Classic Elements. ing on the artist s decision, this prepared information may only be partially exploited. Our method addresses the problem of simultaneous consistency in the 3D object space and in 2D image space while maintaining temporal coherence independently of the mesh structure, but exploiting the very high level of uniformness and isotropy of low-discrepancy sequences, providing also infinite level of detail control. Due to the pipeline requirements, we decompose the process to a 3D processing phase where the algorithm prepares information about all possible strokes, which are interactively decimated or fine-tuned during the post-production phase. 3 The Proposed Rendering System Our rendering system consists of components that are associated with the 3D rendering and the post-production phases of the production pipeline. During 3D rendering, we can use the geometric representation of the rendering software, which assumes that surfaces are already tessellated to 3D triangle meshes. Figure 1 depicts the classical rendering pipeline and our NPR pipeline. As the complete 3D geometry is available in the 3D rendering process, all processing steps that exploit this information must be executed here. Strokes that may visualize contours and hatches are computed at this stage and are output in special files or layers to the post-production phase. Strokes are represented as the control points of Catmull-Rom splines. The 3D rendering process cannot rely on artistic decisions since they happen later in the postproduction phase. Thus, during 3D rendering we create all possible lines of the maximum lengths, which may be decimated or shortened by the artist. During post-processing, splines are extruded to triangle strips. Artists typically would like to define the texture, width and opacity of the contour lines which can be performed at interactive rates. These lines can even be further refined in various ways. Line width or opacity can be varied according to distance from the camera, an illumination image or any kind of mask image. Lines can also be split and scaled to create short brush lines. Lines can also be split at sharp corners mimicking what, in classical artwork, the artist would draw using separate lines crossing each other. 3.1 Contour Rendering Contours depict the boundary of the visible shape of an object (Figure 2). The silhouette separates the object from its environment, internal contours emphasize internal features. Contours have an important role in presenting the shape of the object. Contours show up in hand-made drawings, pen-and-ink illustrations, and may be added to paintings as well. Contours can be created in the rendered image using edge detection filters. Since we wish to emphasize object 22 CEPIS UPGRADE Vol. XI, No. 6, December 2010 Novática

5 Figure 2: Silhouettes and Internal Contours rendered by our System. boundaries, edge detection is worth executing on the depth image rather than the colour buffer. A significant drawback of this approach is that the resulting contours are rather noisy and are made of independent points and not as a well defined curve. Thus, it is very difficult to assign a specific drawing style to the contour line. Fortunately, contours can also be generated from the 3D object geometry. Silhouette curves contain points where the surface normal is perpendicular to the viewing direction. We can also distinguish interior silhouettes from exterior silhouettes that give the outline of the objects seen from the camera. Ridges and valleys are defined by surface points where the surface normal changes abruptly. To display contours we should extract them from the geometry, which is defined by a triangle mesh. Silhouette contours can be defined as the edges that share a front and a back facing triangle. However this definition leads to a collection of line segments that should be combined with a set of heuristics into long smooth paths. Another solution is to reconstruct the surface normal of the original, not tessellated surface from the discrete surface representation. First, the normal vector is estimated at the triangle vertices as a weighted average of the triangle normals. We use the angle of the triangles at this vertex as weights. Then, the reconstructed normal inside the triangles is obtained as the bilinear interpolation of the reconstructed vertex normals. Taking the reconstructed normal, first we search for silhouette points on the edges of the triangles by checking when the normal becomes perpendicular to the viewing direction (the zero crossing of the scalar product of these two vectors). As these lines will enter and exit at two of the edges of a triangle, continuous silhouette paths can be easily extracted. We can smooth these lines to better approximate the original surface. Ridge and valley curves are extracted similarly; they are defined by the zero crossing of the derivative of the surface curvatures. Contours found so far must be processed further and cut those parts that are not visible from the camera due to occlusions or being outside of the camera frustum. To execute these operations, curves are vectorized and clipped onto the view frustum, then rasterized and the visibility is checked in every pixel. We can use depth buffer test or object and face ID test for visibility checking. If the curve turns out to be invisible in a pixel, the curve is cut here. 3.2 Hatching The goal of hatching is to mimic a gray level shade by the density of strokes, where each of them is drawn to roughly follow the principal curvature directions or to enclose approximately the same angle with them. Mimicking the tone means that in any neighborhood the percentage of pixels that are covered by strokes should be proportional to the target shade, i.e. the density or frequency of lines is modulated by the target shade. From mathematical point of view, this task is a frequency modulation problem where the input is the target shade and the output is a sequence of stroke primitives. In order to maintain the consistency of hatching with Figure 3: Hatch Lines with Different Textures and Styles. Novática CEPIS UPGRADE Vol. XI, No. 6, December

6 Figure 4: Strip Generation for a Catmull-Rom Spline. the 3D geometry, we attach sample points called seeds to surfaces given in the local modeling system. As we place seeds on surfaces before transformations and deformation due to, for example, character animation, the seed is always on the very same point of the object and does not swim on the surface. To obtain uniform and isotropic distribution, we find an area preserving mapping that maps the surface to a unit square and define sample points in the square as a Halton low-discrepancy sequence. Note that this sequence has the property that it can be stopped at any sample number, the already generated samples will uniformly fill up the square, and consequently the 3D surface as well. This provides us with fine and infinite level-of-detail control, so new samples can always be added without modifying the previously generated samples. Seeds are transformed with the object and each seed becomes a point of a tentative hatch line. To obtain the hatch line, the local curvatures are computed and we step into the curvature direction, i.e. we integrate the direction field. The visibility of the hatch curves is determined, similarly to the visibility calculation of contour curves, and finally seeds and their associated hatch curves are mapped onto the screen. Note that due to surface deformation, perspective shortening and occlusions, the seed points will not be uniform anymore in screen space. Thus, we reject a subset of seed points to restore uniform density in image space. Rejection is just the setting of the maximum index of the Halton sequence above which the seeds are omitted in a certain subset of the screen. In the last step, we ignore further samples to convert the screen space uniform distribution to a distribution defined by the target shade. The reason of not executing rejection sampling in a single step and obtaining the screen space samples mimicking the target shade directly from the non uniform samples projected on screen is that the distortion due to animation and perspective shortening must be handled separately from the mimicking of the target illumination since the former is independent of the artistic style, but how intensities are handled should be interactively controllable. In the post-production framework, we take hatch lines associated with the remaining seed points and draw them on screen (Figure 3). 3.3 Textured Line Drawing Both contour lines and hatch lines arrive at the postproduction stage as Carmull-Rom splines defined by their control points. Here, these 2D curves are drawn to provide a hand-crafted look while allowing interactive artist control. Hand-drawn curves have roughly constant width on the image independently of the object s distance from the virtual camera. The width and the image of a stroke depend on the artist s device used to draw this stroke. Hand-drawn strokes are not geometrically exact, they have inherent randomness. In order to present naturally looking line art, curves are fattened to polygons and rendered as textured polygon strips where the texture simulates brush or pencil strokes. Textured line drawing consists of two basic tasks. We have to fit a triangle strip to the curve that represents the path of the stroke, and texture coordinates need to be calculated for the triangles. Both tasks are performed in image space. The goal of stripification is to create a triangle strip that fits to a curve. First, the curve is decomposed to shorter segments of similar arc length. Computing the derivative, Figure 5: Three Different Textures simulating Brushes and a Curve Textured with them. 24 CEPIS UPGRADE Vol. XI, No. 6, December 2010 Novática

7 Figure 6: The Minimax Filter. Figure 7: Our Painterly Rendering System controlled differently in the Three Focus Layers. Figure 8: The Image of a Village (left), and the Results of the Application of the Painterly Rendering Module using Two Different Settings (middle and right) i.e. the tangent vector at the end points of the segment and rotating them by 90 degrees, we obtain the normal vectors of the curve. Translating the end points in the direction of the normals, we can calculate the vertices of the enclosing polygon strip (Figure 4). The u texture coordinate for the resulting strip is computed from the arc length. The v coordinate is 0 or 1 depending on whether the vertex is the translation of the curve point by the normal or in an opposite direction. 3.4 Painterly Rendering Our painterly rendering or image abstraction algorithm is based on a filter that is often referred to as Minimax. The main idea of this filter is to colour a pixel by either the minimal or maximal colour value of pixels in its neighborhood. If the maximum is selected, the resulting picture is brighter. Using the minimum, the image gets a darker tone. In either case, the amount of blur introduced by the filter depends on the size of the considered neighborhood. Our method generalizes the original Minimax filter in three ways: we make the size of the neighborhood varying, allow the user to set up a brush to define the shape of the neighborhood, i.e. a mask selecting which pixels belong to the neighborhood, and finally add an automatic control to the brush size using the distance from the camera, which can be read from the depth buffer generated by the 3D rendering process anyway. We use this information to allow the user to separate the scene into three layers: the foreground, the midground, and the background. To prevent unwanted sharp edges arising from the change in the neighborhood size at layer boundaries, a smoothness parameter can be specified, to control the neighborhood size blending from one layer to another. This layering of the input images allows the artist to blur all parts of the scene that are not in focus simulating an artistic depth of field effect. Novática CEPIS UPGRADE Vol. XI, No. 6, December

8 Figure 9: Results obtained with Different Settings of the Contour Generation and Hatch Rendering Systems. 4 Results The discussed methods have been integrated into the movie rendering pipeline of Maya modeling tool, RenderMan 3D rendering software, and Nuke post-processing program. Figure 9 shows the images of Venus with contours and different hatching styles. Figure 10 is a Moria scene rendered with hatching and contouring only. Both shadows and tones are expressed by hatch lines. Our application is used to create a 3D fantasy movie of NPR artistic style. The "main actor" of the movie and his shield are shown by Figure 11 and his village by Figure 12. These models and scenes are the courtesy of Lichthof Productions. Figure 11: The Main Actor of the Rendered Movie and his Shield. Figure 10: Moria illustrated with Hatching and Contouring. 5 Conclusions This paper presented an NPR system and its application in the movie rendering pipeline. Our system implements three NPR effects, contouring, hatching, and painterly rendering. The developed algorithms guarantee that these 26 CEPIS UPGRADE Vol. XI, No. 6, December 2010 Novática

9 Figure 12: The Village Scene with Hatching + Contouring and also with Image Abstraction. effects remain consistent with the animation and the deformation of the 3D geometry, avoid flickering, while still maintaining the look of hand drawn images. This project also proves that NPR effects can be made flexible enough to allow the artist to express his own ideas and produce images meeting his expectations. Acknowledgements This work has been supported in Hungary by OTKA K , and by the scientific program of the "Development of quality-oriented and harmonized R+D+I strategy and functional model at BME" (TÁMOP-4.2.1/B-09/1/KMR ). References [1] L. Kovács, T. Szirányi. Creating Video Animations Combining Stochastic Paintbrush Transformation And Motion Detection. 16th ICPR, Vol-II. pp , IEEE & IAPR, [2] J.E. Kyprianidis, H. Kang, J. Döllner. Image and Video Abstraction by Anisotropic Kuhawara Filtering. Computer Graphics Forum (Pacific Graphics 09), Vol. 29, No. 7, [3] B.J. Meier. Painterly rendering for animation. SIGGRAPH 96. (1996), ACM, pp [4] D. DeCarlo, A. Finkelstein, S. Rusinkiewicz, A. Santella. Suggestive contours for conveying shape. ACM Transactions of Graphics. Vol. 22, No. 3 (2003), pp [5] G. Winkenback, D.H. Salesin. Computer generated penand-ink illustration. SIGGRAPH 94 (1994), pp [6] A. Girschick, V. Interrante, S. Haker, T. Lemoine. Line direction matters: an argument for the use of principal directions in 3D line drawings. International Symposium on Non-Photorealistic Animation and Rendering 2000 (2000), pp [7] S. Rusinkiewicz. Estimating curvatures and their derivatives on triangle meshes. Second International Symposium on 3D Data Processing, Visualization, and Transmission (3DPVT) (2004). [8] O. Deussen, T. Strothotte. Computer-generated penand-ink illustration of trees. SIGGRAPH 00, (2000), pp [9] D. Cornish, A. Rowan, D. Luebke. View dependent particles for interactive non-photorealistic rendering. Graphics interface 2001 (2001), pp [10] R.D. Kalnins, L. Markosian, B.J. Meier, M.A. Kowalski, J.C. Lee, P.L. Davidson, M. Webb, J.F. Hughes, A. Finkelstein. WYSIWYG NPR: drawing strokes directly on 3D models. SIGGRAPH 02 (2002), pp Novática CEPIS UPGRADE Vol. XI, No. 6, December