Computational Aesthetics in Graphics, Visualization, and Imaging (2007) D. W. Cunningham, G. Meyer, L. Neumann (Editors) An Information-Theoretic Ambient Occlusion Francisco González, Mateu Sbert, and Miquel Feixas Institut d’Informàtica i Aplicacions, Universitat de Girona, Spain Abstract Ambient occlusion is a powerful technique that mimics indirect global illumination at a fraction of the cost. In this paper, we introduce a new ambient occlusion technique based on information-theoretic concepts. A viewpoint quality measure is first defined using the concept of mutual information of the channel formed between a set of viewpoints and the polygons of an object. By reversing this channel we can speak of the mutual information of a polygon with respect to all viewpoints. From this polygonal information we represent a kind of ambient occlusion, which is dependent on the importance assigned to each viewpoint and helps to enhance features such as salient parts. Further, the assignation of color to each viewpoint combined with the polygonal information produces a nice visualization of the object. Examples are given with coloroid palettes and non-photorealistic rendering. Categories and Subject Descriptors (according to ACM CCS): I.3.3 [Computing Methodologies]: Computer Graph- icsPicture/Image Generation; 1. Introduction Ambient occlusion is a family of techniques that allow to im- itate the indirect illumination part of global illumination with a very cheap cost. Obscurances [ZIK98] was the first of such techniques, introduced in the context of computer games for fast editing purposes and then used in production rendering. Indirect illumination is decoupled from direct one, and some additional features include for instance color bleeding for further realism [MSC03]. In this paper we define a new ambient occlusion technique based on a viewpoint information-theoretic framework. In this approach, each polygon of an object shares information with the set of visible viewpoints. This shared (or mutual) information is a descriptor of the quality of that visibility and provides a natural ambient occlusion value. In computer graphics, several viewpoint quality measures have been applied in areas such as scene understanding [PB96, VFSH01, PPB * 05], scene ex- ploration [AVF04, SPT06], and volume visualiza- tion [BS05, TFTN05, VFSG06, JS06]. In [FSG07], a new viewpoint quality measure based on mutual infor- mation has been introduced from an information channel constructed between the set of viewpoints and the polygons of the object (Section 2). This measure has also been applied to select the best views in volume visualization [VFSG06]. In this paper, from the reversion of the viewpoint channel, the information associated with each polygon, which we call polygonal mutual information, is defined and used to obtain a kind of ambient occlusion (Section 3). Then, we show how the importance assigned to each viewpoint helps us to enhance characteristics such as the most salient parts and how the polygonal mutual information can be combined with the color assigned to each viewpoint to produce non-photorealistic visualizations (Section 4). 2. Background and Related Work In this section we review some basic concepts of information theory, viewpoint selection and ambient occlusion. 2.1. Information-Theoretic Concepts Let X be a finite set, let X be a random variable taking val- ues x in X with distribution p(x)= Pr[X = x]. Likewise, let Y be a random variable taking values y in Y . An infor- mation channel between two random variables (input X and output Y ) is characterized by a probability transition matrix (conditional probabilities) which determines the output dis- tribution given the input. The Shannon entropy H(X ) of a c  The Eurographics Association 2007.