Immersive Visual Information Mining for Exploring the Content of EO Archives

The amount of collected earth observation data is increasing intensively in order of several Terabytes of data a day. Simultaneously, new trends for exploration and information retrieval are highly demanded. Because recent proposed methods to explore EO data are based on the Image Information Mining IIM approach in which image features extraction, data reduction and labelling are the main steps, developing a new process chain, mainly based on human interaction, might be a promising solution. More precisely, human interacts with features in order to have an active learning system. The focus of this article is based on Immersive Visual Information Mining in which features/images are visualized and modified in an interactive immersive 3-D virtual environment (namely, CAVE) in order to change the learning process and eventually improve its performance. As the first step, the contents of images are extracted and represented by feature descriptors. A library of specific descriptors for multispectral and SAR is used: it comprises spectral-SIFT, spectral-WLD, color-histogram, color-SIFT and color-WLD. Thus, the whole archive is represented in the n-dimensional space of the extracted features, each patch being a point. In optical EO images, color-histogram can be extracted by concatenating the local histograms of colors for the three, RGB, channels. To build the two latter feature descriptors (color-SIFT and color-WLD) the spectral descriptors are applied individually to each color channel, and then they are concatenated to generate the final feature vectors. Each particular feature descriptor represents a particular aspect of the images.