Enhancement of Moment Based Painterly Rendering Using Connected Components M. Obaid, R. Mukundan, T. Bell Department of Computer Science and Software Engineering University Of Canterbury Christchurch, New Zealand. {mho33@student.canterbury.ac.nz} Abstract Moment functions have been used recently to compute stroke parameters for painterly rendering applications. The technique is based on the estimation of geometric features of the intensity distribution in small windowed images to obtain the brush size, colour and direction. This paper proposes an improvement of this method, by additionally extracting the connected components so that adjacent regions of similar colour are grouped for generating large and noticeable brush stroke images. An iterative coarse-to-fine rendering algorithm is used for painting regions of varying colour frequencies. Performance improvements over the existing technique are discussed with several examples. Keywords--- Painterly rendering, non-photorealistic rendering, geometric moments, connected component image. 1. Introduction Artistic rendering has become an important research area in Computer Graphics because of the many challenges posed by the general problem of stylized approximation of an image. This field is inspired by various artistic styles such as paintings [1, 2], drawings [3, 4], animated cartoons [5, 6] and technical illustrations [7, 8]. These artistic styles can be grouped into two categories according to their input data: 3D object-based [9], which takes 3D model of a scene as their input; and 2D image-based [10], which takes 2D images as their input. Interactive artistic rendering techniques are commonly used in digital painting systems and provide the user with a wide range of options and tools. Haeberli [11] introduced such a system that allows the user to place brush strokes manually on a canvas. Each brush stroke is described by its location, colour, size, direction and shape. Haeberli also proposed an automatic way of controlling the brush stroke orientation by using the gradient data of the source image. Non-Interactive methods are a lot more complex to design and implement, as the system needs to extract shape features automatically from input data, and then map these features to the most appropriate brush-stroke parameters for stylized rendering. Hertzmann [1] proposed a method for automatically painting brush strokes using spline curves. Painting is done on several layers, where larger stokes are used in lower layers and thinner strokes in upper layers. The intensity gradient of the image controls the orientation of the spline curves. Geometric moments [12] are popular shape descriptors for image analysis. They have been employed in non-interactive painterly rendering applications for estimating stroke parameters based on local intensity distributions. In this paper, we present a method that aims to improve upon the previous work using this approach by Shiraishi and Yamaguchi [10]. Shiraishi’s method first computes the brush stroke locations from the source image, and then the stroke parameters such as the location, length, width, and angle are calculated using image moment functions. Strokes are then painted using alpha blending in the order of largest to smallest. Our method extracts the connected components from the image to identify the shape of larger brush strokes. This greatly enhances the painterly appearance of the image. Nehab and Velho [13, 18] extended the work of Shiraishi and Yamagushi by using a multi-resolution technique and introducing parametrized stroke positions image. They also proposed a stroke placement method with the addition of a stroke normalization and a dithering method for stroke positioning using a variance filter and blue-noise dithering. This paper is organised as follows. Section 2 describes the use of geometric moment functions as shape descriptors. Section 3 gives an overview of the moment based painterly rendering algorithm. Section 4 describes how connected colour components could be easily computed. Section 5 explains how the proposed method uses connected components for generating more expressive strokes for artistic rendering. Section 6 presents some of the results obtained. Finally, section 7 concludes the paper and outlines some future directions in this area.