Stylized Stereoscopic 3D Line Drawings from 3D Images

Abstract

Stereoscopic 3D (S3D) line drawings were introduced by Sir Charles Wheatstone in 1838. S3D line drawings persist today in various art forms, such as comic books. Stereoscopic 3D line drawings may be hand-drawn or generated from 3D meshes using a variety of algorithms. When creating these drawings, emphasis is placed on consistency: ensuring that the object/scene visible in both views matches exactly for a comfortable viewing experience and accurate depiction of depth [Northam et al. 2013]. While producing S3D line drawings from S3D photos has not been studied in depth, several methods do exist. Kim et al. describe a method for producing stylized stereoscopic 3D line drawings from S3D photographs [Kim et al. 2012]. Their paper applies Canny edge detection to the edge tangent field [Kang et al. 2007] of the left stereo image and warps the discovered edges to the right image using the disparity map. However, the rendered lines are from all edges that can be found in the actual image, including object contours as well as texture or lighting contours. By contrast, a hand-drawn stereoscopic 3D line drawing would be likely to include only object contours and creases. In previous work, we explored the stylization of S3D images by decomposing an image into a set of disparity layers [Northam et al. 2013]. However, that would be ineffective here because while applying the Canny edge detector to the disparity mapwould isolate object contours from texture or lighting contours, the layers would only contain pixels of a single disparity. Hence, there would be no edges to find in each layer. We present a method to produce stylized stereoscopic 3D line drawings from 3D photos that only depicts object contours similar to traditional line drawings. Since contours alone can be insufficient to communicate 3D shape, we also provide the option of adding shading to our drawings to clarify shape and enhance the perception of depth.