Due to the limitations of materials and manufacturing processes, stripe non-uniformity is commonly present in infrared images, which seriously affects the imaging effect of the image and subsequently interferes with subsequent target recognition, detection, and other work. The classic least mean square error (LMS) algorithm can suppress stripe non-uniformity to a certain extent, but its scene adaptability is poor, and there are trailing and “ghost” phenomena. This article proposes an improved least mean square error (LMS) adaptive filtering algorithm for image processing, which utilizes bilateral filtering and steepest descent method to quickly obtain accurate correction parameters. The correction results calculated from the previous frame are used as the initial input values for the following frame, improving the accuracy of the algorithm. At the same time, the algorithm also adds an edge detection module to preserve image details. The article uses real infrared images of non cooled detectors in different scenarios, and compares the algorithm proposed in this paper with the classic LMS algorithm from both subjective and objective aspects. The results show that the algorithm proposed in this paper can effectively protect image details and has good scene adaptability.