Deep learning to classify difference image for image change detection

Jiaojiao Zhao, Maoguo Gong, Jia Liu, Licheng Jiao

Abstract: Image change detection is a process to analyze multi-temproal images of the same scene for identifying the changes that have occurred. In this paper, we propose a novel difference image analysis approach based on deep neural networks for image change detection problems. The deep neural network learning algorithm for classification includes unsupervised feature learning and supervised fine-tuning. Some samples with the labels of high accuracy obtained by a pre-classification are used for fine-tuning. Since a deep neural network can learn complicated functions that can represent high-level abstractions, it can obtain satisfactory results. Theoretical analysis and experiment results on real datasets show that the proposed method outperforms some other methods.