数字全变分滤波在红外图像序列非均匀校正中的应用

Application of digital total variation filter to nonuniformity correction in infrared image sequences

红外焦平面阵列上的空间非均匀性,即固定图案噪声,严重影响了红外系统的成像质量。为此提出了一种新的非均匀校正算法,它借鉴了图像处理中的图像去噪思想,把数字全变分滤波器引入非均匀校正。鉴于数字全变分滤波的执行过程一般都是自我迭代,而考虑到校正算法的实时性要求,无法对单帧图像实施多次处理。故而将以前的空间迭代形式转化为时间递推,以达到逐帧去除固定图案噪声的目的。使用仿真和真实红外图像序列对算法性能进行验证,实验结果表明:该算法从第15帧开始就能够达到较好的校正水平,并且在未优化条件下,每秒钟处理30帧图像仅需300M个指令周期。因此具有收敛速度快、计算复杂度低等优点。

Spatial nonuniformity, known as fixed-pattern noise, in infrared focal plane arrays causes degraded imaging quality. In this paper, a new nonuniformity correction algorithm was presented, which was inspired by the idea of image denoising in image processing and digital total variation filter（DTVF） was introduced to nonuniformity correction. Considering that DTVF was performed iteratively. However, to meet the requirement of real-time processing, our algorithm could not repeatedly handle an image many times. Therefore, the spatially iterative operation was converted into a temporally recursive one and images were corrected frame by frame. Both simulated and real infrared image sequences were emploied to test the algorithmic performance. Experimental results indicate that the proposed algorithm has reached a higher correction level since frame 15, and processing 30 frames per second only consumes 300M instruction cycles without any optimization. So its advantages lie in high convergent speed and lowcomputational comp｜exity and et al.