The traditional neural network method is faced with the problem of gradient step size selection when dealing with nonuniformcorrection of infrared images. When the gradient step size is large, it is easy to cause gradient divergence, andwhen the gradient step size is small, it is difficult to obtain convergence.At the same time, the algorithm is also prone toghosting or image blurring. Aiming at this problem, this paper proposes an infrared image non-uniform correctionmethod based on adaptive forgetting factor recursive least squares method. Firstly, this paper deduces the least squaresmethod into the form of incremental calculation, and introduces it into the calculation of the offset and gain of nonuniformcorrection, so that it can train the infrared image frame by frame. At the same time, this paper considers theproblem that the background of the previous frame is learned to generate ghosts in the process of image from long-termstill to sudden change, and the calculation of forgetting factor is introduced. And this paper uses local structuralsimilarity index (SSIM) to calculate the forgetting factor. The experimental results show that the iterative step size of theproposed method can be calculated adaptively, without manual adjustment, and can effect overcome the ghost problem.Compared with the traditional neural network method and time domain high-pass filtering method, the algorithm of thispaper is the best.
展开▼