Strip Noise Removal of Infrared Images Acquired by Trackside Based on Noise Adaptive Modification

Infrared thermal imaging plays an important role in condition monitoring and damage detection in the industrial field. However, due to the non-uniformity of the response of the infrared detector pixel, there is stripe noise in the infrared image, which affects the image quality, and then causes problems in the later image processing, recognition and classification. In order to solve the problem of stripe noise in infrared images of trains acquired by the trackside, this article fully utilizes the prior knowledge of stripe noise and regards noise as a “special” texture. In the actual situation that no noise-free reference images can be obtained, a noise correction method based on adaptive modification of stripe noise is proposed. The initial stripe noise is separated by one-dimensional guided filtering (1D-GF), and then the interval gradient is used to adaptively modified the separated initial stripe noise, and the final correction result is obtained through the modified stripe noise. Through qualitative and quantitative comparative analysis of images of trains acquired by the trackside and infrared images of other datasets, the results show that the proposed method has a better stripe non-uniformity correction effect and better edge detail retention ability compared to several existing correction methods. In addition, the proposed algorithm is robust to different types of stripe non-uniform noise.