Abstract: There are three major problems in respect of image de-noising：the first is the contradiction between noise depression and detail preservation；the second is the contradiction between the real time request and the algorithm complexity；the third one is the contradiction between the idealized statistical model and the variety of the real noise. As different types of filter templates are adaptable to different noise models and different sizes of filter templates retain varying degrees of detail, the state equation and observation equation are established for different types and sizes of filter templates, and a nonlinear model is approximated a linear model by the use of Extended Kalman Filter（EKF） theory. Image is de-noised depending on observed value of different templates. The simulated result shows that Kalman filter based on template iteration has advantages such as less calculation, more rapid speed and higher quality, especially adaptation to mixed noise compared with other filter algorithms.

Key words: de-noising, template iteration, nonlinear model, Extended Kalman Filter（EKF）

摘要： 图像去噪面临的问题主要有三点：一是噪声抑制和细节保留之间的矛盾；二是实时性与算法复杂度之间的矛盾；三是理想化噪声统计模型与噪声多样性之间的矛盾。由于不同类型滤波模板对不同噪声模型适应性不同，不同尺寸的滤波模板对细节保留程度不同，为此，建立不同类型、不同尺寸的滤波模板的状态方程和观测方程，利用推广的卡尔曼滤波理论将非线性模型近似线性化，利用各模板的观测值，进行卡尔曼滤波。

关键词: 噪声抑制, 模板迭代, 非线性模型, 推广卡尔曼滤波