EG UK Theory and Practice of Computer Graphics (2008) Ik Soo Lim, Wen Tang (Editors)  The Eurographics Association 2008. A Fast Algorithm for Painterly Rendering on Mobile Devices R. Mukundan, C. Han Department of Computer Science and Software Engineering, University of Canterbury, Christchurch, New Zealand. ____________________________________________________________________________________________________ Abstract With the rapid growth of mobile graphics applications, non-photorealistic rendering algorithms developed particularly for devices with limited processor capabilities have become important in the areas of games design and augmented reality. This paper presents a fast painterly rendering algorithm suitable for implementation on mobile phones. Connected components in an image are identified and stored in an index buffer, using a sequential scan. Most of the subsequent processing is done only on this index buffer that contains one integer value per pixel. The proposed method does not use recursive procedures, complex floating-point computations, or texture processing functions. The painterly rendered effect is produced by suitably modifying the boundary of connected components and highlighting edges using entries from the index buffer. The paper presents the theoretical framework for the algorithm, implementation aspects and results. Categories and Subject Descriptors (according to ACM CCS): I.3.3 [Computer Graphics]: Picture/Image Generation ____________________________________________________________________________________________________ 1. Introduction Non-photorealistic rendering (NPR) styles have been extensively used in games, cartoon animation, and presentation graphics. Painterly rendering algorithms form a subclass of NPR algorithms, where stylistic depictions of two-dimensional images are produced using simulated brush strokes [Her98], [HJO01] or sketches [GG01]. Painterly effects can also be generated using particle systems [Mei96]. With recent advances in the areas of graphics hardware specifications and shader programming, impressive artistic effects could be generated using the GPU [VBT*06]. Painterly rendering algorithms typically use shape feature descriptors to identify, transform and map brush stroke parameters based on regions of similar color [KS05], [ON05]. Recursive connected component labeling methods can be used to obtain regions of nearly the same color value. Shape parameters can them be extracted from these regions using global feature descriptors like moment functions. Such moment based algorithms have been recently proposed and successfully used for painterly rendering [NV02], [SY00]. Elegant methods for automated painterly rendering have also been developed using computer vision techniques for digitally simulating brush strokes [GCS02]. Another computer vision based approach that uses eye-tracking data for painterly rendering can be found in [SD02]. NPR algorithms find interesting applications on mobile devices such as painterly rendering of captured images, mobile games, puzzles and artistic rendering of three- dimensional objects. Another emerging application area for NPR algorithms is mobile Augmented Reality [HS04]. The increasing importance of NPR in the field of mobile graphics motivates us to develop fast and efficient methods suitable for processors with limited power and memory. This paper presents a method that is derived from the moment-based algorithm using connected components [OMB06]. It replaces complex data structures and procedures with simple functions that can be easily evaluated on a mobile phone. For example, recursive functions for connected component labeling have been replaced with iterative procedures involving a sequential scan. The computations of moment functions, dither images and stroke density are not performed. Instead, region