基于粒子滤波的红外焦平面阵列非均匀校正算法

Nonuniformity correction algorithm of infrared focal plane arrays based on particle filters

空间固定图案噪声严重影响红外探测系统的成像质量,并且还会随着外界环境的变化而缓慢漂移。卡尔曼滤波算法可以解决这个问题,但会受到线性高斯模型的限制。提出了一种利用粒子滤波跟踪噪声漂移,进而实现非均匀校正的算法。首先使用粒子采样噪声参数可能的取值状态,然后通过状态空间模型对这些粒子进行预测和更新,最后把粒子状态的加权平均值作为对噪声参数的估计。该算法是对卡尔曼滤波算法的扩展,不受任何模型条件的限制,因此,能够实现对不同情况噪声漂移的有效跟踪。实验结果表明:经此算法校正后的图像的峰值信噪比平均都在37dB以上,而经卡尔曼滤波算法校正后的图像却只有28dB。

The performance of infrared imaging systems is significantly affected by spatial fixed-pattern noise,which is mainly due to nonuniformity in the photo response of the individual detectors for the same irradiance. Furthermore, this noise drifts slowly and randomly with time. Kalman filtering approaches can be used to solve this problem. However, their applications are restricted by the linear Gaussian model. A novel nonuniformity correction algorithm that used particle filter to capture the drift in fixed-pattern noise was developed, and the tracking course could be separated into three stages. First, particles were exploited to sample the probable states of noise parameters. Second, these particles were predicted and updated by the state-space model. Finally, the noise parameters could be estimated by the weighted average of the particle states. The proposed algorithm was an extended version of the Kalman filtering method, which could be suitable for any state-space mode/. Therefore, it had a better tracking ability than the traditional Kalman- filtering methods on condition that the model was nonlinear or non Gaussian. The experimental results indicate that the average peak signal-to-noise ratios of the corrected images are all higher than 37 dB with the proposed algorithm, while only 28 dB with the Kalman filtering method.