基于粒子滤波和模拟退火方法的图像配准

图像配准技术在很多领域都应用广泛，如在医学图像分析、计算机视觉、遥感图像等。在仿射变换模型下，只需知道仿射变换参数即可对图像进行配准。为了准确快速地配准，论文在贝叶斯框架下，提出了一种基于粒子滤波和模拟退火法相结合的图像配准算法。首先通过观测模型得到参数的后验概率分布，利用粒子滤波逐步迭代并更新权值来逼近参数的真实值，再利用重构的概率分布模型获得准确的仿射变换参数，对图像进行配准。在粒子更新阶段利用模拟退火算法改进寻优过程，提高运算效率，使结果快速收敛。最后通过实验验证，该文方法易于实现，有较高的准确性和鲁棒性。

Image Registration Based on Simulated Annealing and Particle Filter

Image registration techniques are widely used in many fields, such as medical image analysis, computer vision. In the affine transformation model, there only needs to know the affine parameters for image registration. For accurate and fast registration, in the frame- work of Bayesian, this paper presents a method to image registration combined with particle filter and simulated annealing. The probility dis- tribution function（PDF） is obtained through the observation model, and the affine transformation model and iteration and updating weights are used to estimate the true value. Reconstructed PDF is used to obtain an accurate estimation of the affine transformation parameters. In the stage of particle update, simulated annealing algorithm is used to improve the optimization process. It improves operational efficiency and converges the results fast. Experimental results show that the method is easy to implement and accurate to estimate the affine parameters. Also it has high accuracy and robustness.