In this paper, a multi-level image representation model is developed and used for multi-spectral remote sensing image retrieval in order to narrow the gap between the low-level feature and high-level semantic. This model consists of an image segmentation part, a feature extraction part and semantic extraction part. The first two parts aim at the extraction of primitive region feature of an image. In these two steps, an improved JSEG algorithm is used to segment the image stored in the database, then spectral feature and texture feature are extracted for each region. In semantic extraction part, the semantic information hidden in different regions of different images is extracted by Bayesian method and expectation maximization (EM) method. At last, positive example and negative example concept is used in image retrieval instead of relevant feedback. Experiment shows that this method not only improves the accuracy of the result but also decreases the complexity of retrieval.

Date Published: 29 December 2008
PDF: 8 pages
Proc. SPIE 7285, International Conference on Earth Observation Data Processing and Analysis (ICEODPA), 72850T (29 December 2008); doi: 10.1117/12.815899

Published in SPIE Proceedings Vol. 7285:
International Conference on Earth Observation Data Processing and Analysis (ICEODPA)
Deren Li; Jianya Gong; Huayi Wu, Editor(s)