为实现红外焦平面阵列(IRFPA)盲元的自适应快速校正, 本文设计了一种基于三梯度阈值检测的快速盲元校正方法。先通过优化L0范式的约束方程消除图中的非均匀条纹, 排除干扰, 再对水平、竖直和对角3个方向进行梯度阈值检测, 找到盲元位置点, 并与中值滤波结果进行点对点匹配, 剔除误检测点, 最后通过局部中值滤波完成盲元校正。依据盲元的缓变特性, 通过合理更新盲元模板实现了单帧和序列图像的快速盲元校正。实验结果表明: 相比于传统方法, 本文方法的信噪比(SNR)和峰值信噪比(PSNR)提高了10~14 dB, 结构相似度指标(SSIM)提高了0.01~0.02, 对图像中的随机和连续盲元的校正效果都很好, 同时运算速度得到了大大提升, 在保证图像质量指标的情况下速度提升了3~10倍, 可以在到实际红外系统中实现动态检测和实时处理。

A fast blind pixel correction method based on three-gradient threshold detection was designed in this paper,to achieve adaptive and fast blind pixel correction of the Infrared Focal Plane Array (IRFPA). First, a constraint equation through optimization of L0 paradigm was used to eliminate the interference of non-uniform stripe; then gradient threshold detection was respectively conducted in the horizontal, vertical and diagonal directions to find out the position of the blind pixel, thus through a point-to-point matching with the median filtering results to remove the inaccurate detecting point; finally, blind pixel correction was completed through local median filtering. In the process, reasonable updating of the blind pixel template, based on slow variation of blind pixel, was carried out to achieve fast blind pixel correction of single-frame and sequential images. The results show that compared with traditional method, signal-to-noise ratio (SNR) and peak signal-to-noise ratio (PSNR) achieved by using this proposed method is improved by 10~14dB, with the structured similarity (SSIM) index improved by 0.01~0.02; both random blind pixel and continuous blind pixel in the images can be effectively corrected, and the computing speed is dramatically improved, with a speed advantage of 3~10 times while ensuring the image quality. Hence, such proposed method can be used to achieve dynamic detection and real-time processing in actual infrared system.