(RS)2 I - Retrieval system for remotely sensed imagery on-board image processing for the NTU Satellite X-Sat

• Main Author: Bretschneider, Timo Rolf.
• Other Authors: School of Computer Engineering
• Format: Research Report
• Published: 2008
• Subjects: DRNTU::Engineering::Computer science and engineering::Computing methodologies::Image processing and computer vision; DRNTU::Engineering::Computer science and engineering::Information systems::Information storage and retrieval
• Online Access: http://hdl.handle.net/10356/2312
• Institution: Nanyang Technological University

In recent years, numerous remote sensing platforms for Earth observation have been developed and together acquire several terabytes of data per day. However, the useful utilisation of the imagery by the user imposes a significant challenge to index and retrieval in terms of effectiveness and efficiency. Many professional remote sensing databases retrieve the satellite images based on their world-oriented information such as the location, the scanner name and the acquisition date etc. However, queries that are not directly related to this type of information cannot be processed straightforwardly. An example is the search for a scene that possesses a similar ground cover characteristic like the query specification. For this purpose a variety of content-based image retrieval (CBIR) techniques have been developed and successfully applied in remote sensing databases to facilitate in particular non-professional users. In one of the most prominent approaches the user simply provides a query image, and then the database retrieves similar scenes according to their individual content described by a-priori automatically extracted low-level features. However, the deep gap between these generally low-level features and the high-level semantic concepts on the user side limits the potential of CBIR techniques. For instance, in the case where a user provides a scene with snow covered ground the database may return images with cloud cover based on the similar spectral appearance. To solve the problem, pre-en-tered semantic annotations can be used. However, this supervised process is expensive and inefficient due the subjectivity of the operator.