SAR Images Retrieval Based on Semantic Classification and Region-Based Similarity Measure for Earth Observation

Abstract

Based on the semantic categorization and region-based similarity measure, a novel synthetic aperture radar (SAR) image retrieval method is proposed in this paper, which is inspired by the existing content-based image retrieval (CBIR) techniques and is oriented toward the Earth observation (EO). First, due to the large sizes of SAR images, new method semantically classifies the land covers in the patch level rather than the pixel level by the classic semisupervised learning (SSL), which could reduce the workload of selecting the representative image patch and decrease the searching space in the similarity calculation component. Furthermore, to overcome the inevitable classification error, our method provides an error recovery scheme, preventing the errors produced in categorization to contaminate the retrieval results. Third, the similarity between two patches is calculated by the improved integrated region matching (IIRM) measure based on the region-based similarity measure, which fails to meet the expectation in SAR images. The proposed method can be embedded into any EO mining systems to help them complete the EO missions. After comparing the method presented in this paper to others, it is evident that our method performs more effectively than others from the CBIR aspect.