基于引导滤波的红外图像非均匀性校正

Infraed image non-uniformity correction based on guided filtering

摘要:

由于制作工艺的限制和器件材料的不均匀性,红外图像在一定程度上存在非均匀性,导致目标探测和识别能力下降,严重的情况下甚至无法探测目标,因此,红外图像必须经过校正才能发挥出红外探测器对温度的高灵敏度性能。基于神经网络的非均匀性校正技术是校正非均匀性的有效方法,但在去除非均匀性噪声的同时,会弱化图像信息边缘,导致图像模糊,甚至出现严重的鬼影。为了改善红外图像的非均匀性校正性能,以神经网络模型为架构基础,利用引导滤波算子作为期望真值模板,替代传统的神经网络模型中的均值滤波模板,同时增加鬼影抑制算法,在去除非均匀性噪声的同时,达到抑制鬼影、边缘保真的效果。实验结果表明,提出的非均匀性校正算法能够在保留图像细节特征、抑制鬼影的同时,很好地校正了红外图像的非均匀性。

Abstract:

Due to the limitation of fabrication process and the non-uniformity of device materials, there is serious non-uniformity in infrared images, which leads to the decline of target detection and recognition ability, and in serious cases, the target can not even be detected. Therefore, the infrared image must be corrected to give full play to the high sensitivity of the infrared detector to temperature. Non-uniformity correction technology based on neural network is an effective method to correct non-uniformity, but while removing non-uniformity noise, it will weaken the edge of image information, thus resulting in image blur and even serious ghosts. In order to improve the non-uniformity correction performance of infrared image, based on the neural network model, this paper used the guide filter operator as the expected truth template to replace the mean filtering template in the traditional neural network model. At the same time, the ghost suppression algorithm was added to achieve the effect of suppressing ghost and edge fidelity while removing the non-uniform noise. The experimental results show that the non-uniformity correction algorithm proposed in this paper can not only retain the image details and suppress the ghost, but also correct the non-uniformity of the infrared image