Segmentation of FLIR images by Hopfield neural network with edge constraint

摘要

A segmentation algorithm of forward-looking infrared (FLIR) images by Hopfield neural network (HNN) with edge constraint is presented. An evaluation criterion based on distinct edge pixels is used to examine the segmentation results by HNN under different initial assignment of probabilities. Thus, the good segmentation result can be achieved by automatically adapting initial assignment of probabilities to reach the optimal or suboptimal solution of the evaluation criterion. To determine appropriate weights of the objective function and the constraint condition in the energy of HNN, a criterion with respect to the constraint condition is proposed. Experimental results with real FLIR images are given.