A filter-based post-processing technique for improving homogeneity of pixel-wise classification data

摘要

Many studies have presented various classification techniques for improving the accuracy of image classification, but heterogeneous classification results, like salt-and-pepper still appear in thematic maps. In this paper, a filter-based post-classification technique, likelihood class filter (LCF), is presented to not only remove heterogeneous classes but also to improve the accuracy of image classification. This paper demonstrates that the classification accuracy can be effectively improved by LCF, which offers the resulting thematic maps of Salinas-A scene, Indian Pines test site and Pavia University scene the optimal overall accuracy (the highest homogeneity index) of 99.81% (0.9716), 92.41% (0.8936) and 92.35% (0.8985), respectively.