CBIR methods in satellite cloud image retrieval

Yan-bing LI,Yuan-xiang LI,Si-Xun HUANG,Jing-qiu ZHAI

DOI: https://doi.org/10.3969/j.issn.1009-3443.2008.02.006

2008-01-01

Abstract:In order to efficiently manage and make use of large-scale satellite cloud image collections, it is useful and essential to build a database and to equip it with retrieval functions, especially content-based retrieval ones. Content-Based Image Retrieval (CBIR) techniques were applied to satellite cloud images. A series of methods were used to extract color, texture and shape features. Color features used were gray level histogram and gray level coherence vector. Texture features were extracted from wavelet subbands using two methods: histogram fitting which used parameters of generalized Gaussian density functions fitting subband histograms as features and partitioned histogram fitting method. Shape features were extracted using a new-proposed deformable model method aiming at cloud objects, called circle unit method. After image distance or similarity rules were given for each method, experiments were made. Results indicate that the application of CBIR in satellite cloud image retrieval is effective, and that specially designed shape method achieves much better performance than common methods due to the use of satellite cloud image domain knowledge, though at the same time more time-consuming.