A GPU Implementation of Non-photorealistic Stained Glass Image Generation using Voronoi Diagrams

Abstract

In this paper, we propose a high quality stained glass image generation. Our proposed method converts automatically the original image into a high quality stained glass image. In the method, we use Voronoi diagrams to reproduce a stained glass such that Voronoi cells and edges represent colored glasses and leads that bond the glasses, respectively. In order to obtain a high quality stained glass image, we merge Voronoi cells and move Voronoi seeds using a local search technique. By varying the size of Voronoi cells in a stained glass image, a stained glass image well-represents local characteristics of the original image. Also, by moving Voronoi seeds to fit Voronoi cells to the original image, we obtain a high quality output image. However, the computing time cost is very high in our proposed method. Therefore, we also propose a GPU implementation to accelerate the computation. Our experimental result shows that the proposed GPU imple- mentation on NVIDIA Tesla V100 attains a speed-up factor of 362 over the sequential CPU implementations.