Human facial illustrations: Creation and psychophysical evaluation Bruce Gooch, Erik Reinhard, Amy Gooch 吉娜烨 2010-09-03 Outline • Author Information • Abstract • Motivation • Black-and-white Illustrations • Black-and-white Caricatures • Validation • Conclusion Bruce Gooch(1/2) • HomePage – http://www.thegooch.org/ • Professional History – 07/2007-Present • President and Manager at Toon-FX INC. – 09/2006-Present • Associate Professor at University of Victoria – 09/2003-Present • Assistant Professor at Northwestern University – 06/2001-06/2003 • National Science Foundation Graduate Research • Research – NPR, Perception and Graphics, Learning and Video Games, Computational aesthetics Bruce Gooch(1/2) • Publications – Book • Bruce Gooch and Amy Gooch, “Non-Photorealistic Rendering”, A.K. Peters Ltd. Publishers, July 2001, ISBN: 1-56881-133-0. – Research Papers • A Gooch, B Gooch, P Shirley, E Cohen, “A non-photorealistic lighting model for automatic technical illustration”, ACM Siggraph, 1998. • E Reinhard, M Adhikhmin, B Gooch, Shirley, P., “Color transfer between images”, IEEE CG&A special issue on Applied Perception, 2001. • B Gooch, PPJ Sloan, A Gooch, P Shirley, R Riesenfeld, “Interactive technical illustration”, I3D ’99. • …… Erik Reinhard • Homepage – http://www.cs.bris.ac.uk/~reinhard/ • 01/2006-Present – Working at the University of Bristol, UK. • Publications – Book • E Reinhard, G Ward, S Pattanaik, P Debevec, “High Dynamic Range Imaging ”, August 2005, ISBN 0-12-585263-0, Hardback. – Research Papers • E Reinhard, M Stark, P Shirley, J Ferwerda, “Photographic tone reproduction for digital images”, ACM Transactions on Graphics , pp. 267-276, 2002. • E Reinhard, M Adhikhmin, B Gooch, Shirley, P., “Color transfer between images”, IEEE CG&A special issue on Applied Perception, 2001. Amy Gooch(1/2) • Homepage – http://webhome.csc.uvic.ca/~agooch/index.html • Experience – Current Assistant Professor, University of Victoria, In British Columbia, Canada – BS in Computer Engineering from the University of Utah – MS in Computer Science from the University of Utah – PhD in Computer Science June 2006 at Northwestern University • Research – – – – – Image Creation and Manipulation Multispectral Computational Photography Computational Aesthetics and Non-Photorealistic Rendering Casual Educational Games Human Perception in Immersive Environments Amy Gooch(2/2) • Books – Mario Coast Sousa, David Ebert, Amy Gooch, Don Stredney. “ComputerGenerated Medical, Technical, and Scientific Illustration.” ACM SIGGRAPH 2005 Course #31, 2005. – Bruce Gooch and Amy Gooch, “Non-Photorealistic Rendering”, A.K. Peters Ltd. Publishers, July 2001, ISBN: 1-56881-133-0. – Stuart Green, David Salesin, Simon Schofield, Aaron Hertzmann, Peter Litwinowicz, Amy Gooch, Cassidy Curtis, Bruce Gooch. “Non-Photorealistic Rendering.” ACM SIGGRAPH 1999 Course #17. • Research Papers – A Gooch, B Gooch, P Shirley, E Cohen, “A non-photorealistic lighting model for automatic technical illustration”, ACM Siggraph, 1998. – B Gooch, PPJ Sloan, A Gooch, P Shirley, R Riesenfeld, “Interactive technical illustration”, I3D ’99. – …… Outline • Author Information • Abstract • Motivation • Black-and-white Illustrations • Black-and-white Caricatures • Validation • Conclusion Abstract • • • • • We present a method for creating black-and-white illustrations from photographs of human faces. In addition an interactive technique is demonstrated for deforming these blackand-white facial illustrations to create caricatures which highlight and exaggerate representative facial features. We evaluate the effectiveness of the resulting images through psychophysical studies to assess accuracy and speed in both recognition and learning tasks. These studies show that the facial illustrations and caricatures generated using our techniques are as effective as photographs in recognition tasks. For the learning task we find that illustrations are learned two times faster than photographs and caricatures are learned one and a half times faster than photographs. Because our techniques produce images that are effective at communicating complex information, they are useful in a number of potential applications, ranging from entertainment and education to low bandwidth telecommunications and psychology research. 中文摘要 • 本文提出了一种从人脸照片生成黑白影像的方法。同时,提出了 一种交互式变形技术将黑白影像转换为突出并夸张了人脸特征的 漫画。 • 本文通过心理物理学调查来评价实验结果在识别和学习速度上的 精确性以评估结果的有效性。这些调查显示用本文的方法生成的 人脸影像和漫画在识别时和照片一样有效。在学习任务上,影像 学起来比照片快两倍,漫画学起来比照片快1.5倍。 • 由于本文的方法产生的图像在复杂信息的交流过程中有效,因而 它们有很多潜在应用,娱乐、教育到低带宽通信以及心理学研究。 Outline • Author Information • Abstract • Motivation • Black-and-white Illustrations • Black-and-white Caricatures • Validation • Conclusion NPR Images’ Advantage • Convey information better – Omitting extraneous detail – Focusing attention on relevant features – Clarifying, simplifying, and disambiguating shape Applications • Suitable for rapid transmission over low bandwidth networks. • Benefit for visual learning applications because learning tasks can be sped up by using resulting images. • Benefit for research in face recoginition. Illustration (Two-tone image) • Remove extraneous information from the photograph while leaving intact the lines and shapes that would be drawn by cartoonists. Caricature • Use a practical interactive technique from observations of a method to train caricature artists. Outline • Author Information • Abstract • Motivation • Black-and-white Illustrations • Black-and-white Caricatures • Validation • Conclusion Brightness Perception Models • Difference of Gaussians – – – – – Blommaert and Martens 1990; Cohen and Grossberg 1984; Gove et al. 1995; Hansen et al. 2000; Pessoa et al. 1995. • Gaussian smoothed Laplacians – Marr and Hildreth 1980; – Marr 1982. • Gabor filters – Jernigan and McLean 1992. Human Brightness Perception Cell properties Human Brightness Perception Human visual system Scale invariance property Neural structures Interpreted at different levels of resolution Assum Controlled by varying receptive field sizes Blommaert Model (1/3) • The receptive fields of the human visual system – Isotropic with respect to brightness perception – Modeled by circularly symmetric Gaussian profiles positions scales center surround Blommaert Model (2/3) Neural response Luminance distribution L is modeled by a center-surround mechanism Mexican hat shape Blommaert Model (3/3) Brightness Illustration Computing Thresh brightness Input luminance Combined output Thresh luminance Brightness Representation Result Thresholded brightness high-pass filtered image followed by thresholding Canny’s edge detector Illustration Results Super-Portrait Creating FFG(Facial feature grid) Exaggerate User-set grid Super Portrait Results 50% anti-caricatures Source image 50% super portraits Outline • Author Information • Abstract • Motivation • Black-and-white Illustrations • Black-and-white Caricatures • Validation • Conclusion Related Work • Exaggerate the differences from an average face. – Brennan 1985; – Redman 1984; – Rhodes et al. 1987. • A second semi-automated caricature generator is based on simplicial complexes. – Akleman et al. 2000 VS. Caricature Generator [Brennan 1985] Brennan 1985 Proposed technique Caricature Creating • Semi-automatic methods to produce super- portraits and caricatures which rely much less on the presence of well-trained users. Caricature Results Outline • Author Information • Abstract • Motivation • Black-and-white Illustrations • Black-and-white Caricatures • Validation • Conclusion Recognition Time Minimum, Maximum, and Mean Time Over Average Subject Data for Each Condition in Each Experiment Learning Speed and Accuracy Minimum, Maximum, and Mean Number of Trial Iterations Outline • Author Information • Abstract • Motivation • Black-and-white Illustrations • Black-and-white Caricatures • Validation • Conclusion Conclusions and Future Work • Conclusions – Applying a modified model of human brightness perception to photographs. – The resulting illustrations are as easy to recognize as photographs, and are faster to learn. • Future work – – – – Using for fMRI experiments Transfer of information in learning tasks. Animating human facial illustrations. Increase the robustness of the algorithms agaist skin blemishes, freckles, or a slight beard. Thank you!