基于卡尔曼滤波的红外图像增强算法

针对红外图像中的非均匀性噪声的去除问题,提出基于卡尔曼滤波的红外图像去噪及增强算法.在Bayes-ian-MAP框架下分析卡尔曼滤波器对去噪问题的适用性.由于成像电路内部温度上升和参数的细微变化,每个像元的固定模式噪声(FPN)在帧间缓慢变化.基于此点,建立暗帧的噪声模型.将卡尔曼滤波器作用于红外暗帧序列,估计出暗帧中每个像元的FPN水平.引入噪声影响因子(NIF)来评估FPN噪声对像元输出信号的影响.根据NIF自适应地选取每个像元的FPN噪声权重.实际带噪图像减去加权FPN噪声,即得到增强图像.将该算法应用于实拍红外图像,用平均灰度梯度(GMG)评估算法的性能.在目标区域,GMG下降了5.1%,说明算法在去噪的同时很好地保留了目标的边缘.而在平滑区域,GMG下降了85.5%.结果表明,该算法在去除非均匀性噪声,提高图像的对比度方面,取得较好的效果.

Enhancement algorithm for infrared images based on Kalman filter

To remove the non-uniformity noise in the infrared image,the denoise and enhancement algorithm based on Kalman filter is proposed.In the framework of Bayesian-MAP inference,the feasibility of Kalman filter was discussed.Due to the inner temperature increment and the circuit parameters’ minor changes of the imaging system,each pixel’s fixed pattern noise(FPN) varies slowly and non-uniformly between frames.Based on this view,the dark frame was modeled.The level of the FPN was estimated by Kalman filter acted on sequential dark frames.The noise influence factor(NIF) was introduced to evaluate the influence of the FPN to the pixel’s output signal.And then,the reasonable weight of each pixel’s FPN was determined adaptively by means of NIF.The non-uniformity of the infrared image can be corrected after weighted subtraction FPN from the real noisy image data on pixels one by one.The algorithm was applied in real infrared images and gray mean grads(GMG) was used to evaluate its performance.In target neighborhood,GMG is reduced by 5.1%.It means that the image is smoothed and the edges are preserved well.At the same time,GMG is reduced by 85.5% in smooth neighborhood.The experiments show that algorithm produces an obvious denoise and enhancive effect in both target and smooth regions.