Super-resolution of blurred infrared images using the blur parameters identification on the neural network

Images acquired from an infrared (IR) sensor typically suffer from poor spatial resolution due to the finite size of the lens that makes up the imaging system and the consequent imposition of the underlying diffraction limits. The lost frequency components beyond the diffraction-limited cutoff make the obtained images blur. Currently there are one kind of image processing schemes referred to as super-resolution algorithms available for solving of this problem, including Bayesian analysis methods, set theoretic methods, and Fourier domain techniques. But an estimate of the blur model parameters is essential in these methods. If incorrect blur parameters are chosen then the super-resolution results will be wrong. This work presents an original solution to the blur parameters identification problem in infrared image super-resolution. A back-propagation(BP) neural network is used for the blur parameters identification. In this method, we consider the modulation transfer function (MTF) of an infrared system as a Gaussian type. Mathematical analysis shows that using back-propagation neural network it is possible to identify the parameters of the Gaussian blur. After blur parameters identification, the image can be restored using several kinds of methods. We choose the Poisson-MAP super-resolution algorithm with Markov constraint(MPMAP) as our restoration method. Experimental results demonstrate that the performance of the MPMAP method using the blur parameters identified by our neural network is superior to other blind image restoration methods.