基于孪生神经网络的异源图像匹配方法

Heterogeneous Image Matching Based on Siamese Neural Network

针对现有图像匹配方法在SAR图像和可见光图像匹配过程中受二者非线性强度差异以及SAR图像散斑噪声的影响,造成二者匹配精度低的问题,提出一种基于孪生神经网络的异源图像匹配方法。该方法通过显著性检测选取待匹配图像块,利用孪生神经网络对选取的SAR图像块和可见光图像块进行相似度度量;使用快速搜索策略遍历搜索区域,以加快匹配速度,并使用集成学习,优化匹配精度。实验结果表明,该方法在匹配成功率和匹配精度上显著优于传统的模板匹配方法。

Aiming at the problem that the existing image matching methods were affected by the large difference in nonlinear intensity between SAR image and visible image and the speckle noise of SAR image,which resulting in the low matching accuracy,a heterometric image matching method based on Siamese neural network was proposed.In this method,the image block was selected to be matched by saliency detection;the similarity was measured between the selected SAR image block and visible light image block by using Siamese neural network;the search area was traversed by using fast search strategy to speed up matching speed;ensemble learning was applied to further optimize matching accuracy.Experimental results showed that the proposed method was superior to traditional template matching method in the success rate and accuracy of matching.