A Study Of Texture Rendering Algorithms In Multi-view Video Generation

Modeling based on multi-view video images needs the information of geometry and surface colors. The surface colors always have been got from texture mapping. Texture technology now has received wide attention for its important academic value and wide application.A texture map is a key component of a geometric model, which has a significant impact on its realism. The majority of the previous works on this problem dealt with accurate geomatric models acquired by an active method such as laser scanning or structured light. However, research and experiments show that low-accuracy geometric models and fine textures can generate better realistic scenes. Image based algorithms rely on image cues. Due to the inherent uncertainty of such cues, the resulting models usually are noisy noisy or biased.For the reasons above, the texture mapping methods always cause seams. Massive texture data is also hard to store and transmit.To deal with these problems, we bring up a surface texture generating method based on multi-view video images. The video rendered by this method can be viewed from any view point. Texture image is also small, which is easy for transmitting.A seamless texture mapping algorithm was proposed since that in the image-based modeling context, color discontinuities at patch boundaries are a crucial issue due to photometric and geometric inaccuracies. With Markov random field which relates the mesh vertices to the viewpoints, a seamless texture map is constructed via barycentric coordinate weight color interpolation. Experiments suggest that the Markov random field energy model can ensure texture map visual quality while minimizes the seam, and the seam can be totally removed with the help of the weight color interpolation.