Modeling of Smoke from a Single View

This paper presents a simple method for modeling smoke from a single view, which could preserve fine realistic look observed from different views. Thin translucent smoke, generated by cigarette, candle, joss stick et al, is the main focus of this paper. The proposed method initially computes the smoke intensity from the input key frame image, then partitions the smoke into multiple segments. For each segment of smoke, the principal direction is calculated by principal component analysis, and two basic functions are generated. Depth of each pixel in the image is estimated with the basic functions. Then, three-dimensional density distributions are constructed by referring to the intensity and depth. Finally, smoke density distributions at key frames are used to generate animated smoke. Experiment results indicate that our method can synthesize visually realistic smoke from a single view with low computational cost.