基于神经优化的最大熵图像重建算法

Maximum entropy image reconstruction based on neural network optimization

提出了一种基于Hopfield神经网络(HopfieldNeuralNetwork,简称HNN)优化的图像重建算法.将图像重建问题转化为HNN优化问题,取重建图像熵函数最大以及原始投影与再投影之间的误差平方和最小作为图像重建的优化目标,作为能量函数构造连续型HNN模型,由HNN能量函数极小化可得到重建问题的优化解.这种方法具有简单易行、计算量小、收敛快、便于并行计算等特点.对照ART算法,用计算机模拟产生的无噪声投影数据检验新算法。

Presents a solution algorithm for the image reconstruction problem based on the Hopfield neural network optimization. We viewed image reconstruction from projections as an HNN energy minimization problem by selecting two criteria as the optimization objection of reconstruction problem, which is maximum of image entropy and minimization of squared error between the original projection data and reprojection data due to the reconstructed image. We designed a Hopfield neural network in continuous work\|mode by mapping the objective function onto the energy function. We can find an optimal solution by minimizing the energy function. This offers the advantages: simplicity of the calculation, less computation, fast convergence and suit for parallel processing. Comparisons of the new algorithm to ART algorithm are carried out using computer\|generated nois\|free projections. The results show that the proposed method is well.