Coast to Coast Seminar Series: Live from Halifax "Artistic Image Processing"

Abstract

Non-photorealistic or 'artistic' rendering is a branch of computer graphics that aims to mimic styles of art algorithmically. Non-photorealistic rendering (NPR) takes its inspiration from historic works created in the natural media of painting, drawing and illustration, and can be made to operate directly on 3D geometry or on existing 2D images. Given the complexity of the task, NPR systems typically focus on the replication of a single artistic style. In this talk I discuss two techniques for artistic image processing, both of which have recently appeared in IEEE Transactions on Visualization and Computer Graphics.

The first of which, entitled "Mixed Media Painting and Portraiture" presents a technique to transform digital images into renderings that approximate the appearance of mixed media artwork, which incorporates two or more traditional visual media. This is achieved by first separating an input image into distinct regions based on the detail present in the image. Each region is then processed independently with a user-selected NPR filter. This allows the user to treat highly detailed regions differently from regions of low frequency content. The separately processed regions are then blended in the gradient domain. In addition, the work is extended to the rendering of mixed media portraits. Portraits pose unique challenges that we address with a method of segmentation based on a composite of face detection and image detail. The approach offers the user a great deal of flexibility over the end result, while at the same time requiring very little input.

The second approach, "Image-Based Stained Glass", attempts to simulate the appearance of stained glass artwork. A stained glass window possesses a distinctive style partly due to the unique color ranges produced through the interaction of color enamels, glass and light. The imposition of lead calmes further separates the appearance of stained glass from other mediums. To simulate this, a novel approach has been developed which involves image warping, segmentation, querying, and colorization along with texture synthesis. In the method, a given input image is first segmented. Each segment is subsequently transformed to match real segments of stained glass queried from a database of image exemplars. By using real sources of stained glass, the method produces high quality results and, again, requires only modest amounts of user interaction.