An Algorithm for Natural Phenomena Simulation Based on Video Texture Synthesis

This paper proposed an improved video-based texture synthesis approach for simulating realistic natural phenomena. The algorithm measures similarity of every two frames to determine a preliminary playing sequence and then divide the preliminary playing sequence into several subsequences with different frame-to-frame delay so as to remove visual discontinuities. For removing the abrupt change from the last frame to the first frame in the video loop, at first our algorithm determines the cycle of the preliminary playing sequence. Then the algorithm chooses several frames from the preliminary playing sequence based on the cycle and divides every single chosen frame or every single chosen subsequence into patches, which depends on the cycle of video. At last the algorithm re-organizes these patches into a final playing sequence. The experiment results indicate that our algorithm is easy to synthesize realistic effects of natural surrounding video.

I. Introduction

In the past few years, computer-based natural phenomena portrayal has attracted much interest of researchers for its wide applications in the real world. Video and image can be used to record phenomena in the nature. And yet, there are many phenomena can not adequately captured by a single static photo. Though the video is considered as the best media on this problem, it has its own drawbacks. The video lacks the “timeless” quality of the photograph. In order to solve the problem, Arno Schodl proposed a new type of medium in 2000, called a video texture, which has qualities between those of photograph and a video. A video texture provides a continuous, infinitely varying stream of video frames. Moreover, in some applications, such as making special effects in movies and video, video textures can be used to instead of very expensively real scene capture.