| 摘要 |
An image processing method comprises: selecting an instrument characteristic matrix P(i, j); for matrix data d(i) of each frame of an input image, calculating a convolution of d(i) and pT(i, j) as a re-convolution matrix c(i); calculating a normalized re-convolution matrix cx(i) of P(i, j) and d(i), setting a threshold fm, determining whether each point of cx(i) is greater than fm, and if yes, determining that the point is a data mutation point or a data point where data is far greater than a background, and repeating the foregoing step until all data points in cx(i) are less than fm; at this time, obtaining matrix data in which data mutation points and data points where data is far greater than the background are removed, and recording the data matrix as db(k); calculating a convolution of db(k) and p(i, j), and recording the convolution as cb(i); restoring a data background through a Gauss-Seidel iterative or Richardson-Lucy iterative algorithm; and repeating the foregoing step until a convergence result is obtained and used as background data after reconstruction. In an image obtained according to this image processing method, data information is restored to a maximum degree, and by fully using priori knowledge about the instrument and target, image restoration and extraction of specific information are implemented in a large dynamic range. |
