基于贝叶斯最大化后验估计方法的图片合成模型研究

Image Synthesis Model Based on Bayesian Estimation Method of Posterior Maximum

在图像处理应用中,常常需要根据一些列相关的输入图片生成一张新的图片。现有的研究大都设定一些启发式规则用于图片的合成过程。为了提高图片合成的性能,提出了一种基于改进的贝叶斯方法的图片合成模型。在给定理想的图片合成模型后,对传感器误差和图片误差进行了分析。由于图片误差和几何误差之间是相关的,因此分析了它们之间的关系。在根据已有数据对模型进行后验估计时,通过最小化能量来得到模型的先验参数。在目标函数的优化过程中,基于现有研究通过重新赋权值的迭代方法进行优化问题的求解。最后,通过大量的实验表明,所提出的图片合成模型与相关方法相比具有更好的图片合成和渲染效果。

In the area of image processing, it usually needs to generate a novel image via a series of related input images. Most of current researches set some heuristic rules in the process of image synthesis. In order to improve the efficiency of image synthesis, this paper proposed a Bayesian based image synthesis model. Given the ideal image synthesis model, we analyzed the errors of sensors and images. As the error between image and geometric is related, we further analyzed their relationship. While doing posterior estimation with given image data,we got the prior parameters of the model by minimizing energy. In the process of optimizing the target function, we applied the re-weighted iterative method based on related works. The experiments show that the proposed image synthesis model has better performance in image synthesis and rendering than related works.