A GPU Implementation of Watercolor Painting Image Generation

Stroke-based rendering is a technique to generate images with the painting effect of an actual brush stroke by repeating a drawing, called a stroke, many times. In this paper, we propose a watercolor image generation with stroke-based rendering. We use a physical model that simulates a watercolor painting on paper by computing the movement of water and pigment. This method requires a large number of strokes to be drawn one by one, and the computational cost of watercolor simulation for each stroke drawn is also very high. Therefore, to reduce the generation time, we propose a parallel algorithm for drawing multiple strokes simultaneously. The idea of the parallel algorithm is to generate multiple strokes independently. However, since some of the generated strokes have overlaps, such strokes cannot be drawn simultaneously. In our approach, we find non-overlapping strokes by reducing this problem to the independent point set problem and solving it instead. Furthermore, we implement this parallel algorithm on the GPU. Experimental results show that the proposed GPU implementation on NVIDIA GeForce RTX 3090 attains a speed-up factor of up to 75 over a sequential execution on the CPU.