Urban area understanding based on compression methods

In this paper, we present a comparative evaluation of two content-based image retrieval systems, the first one based on texture feature extraction methods and machine learning algorithms and the second one based on compression methods and similarity metrics. The evaluation is carried out using high resolution optical and SAR data. The test data set is composed of 4000 tiles, with 64×64 pixel size. Those tiles were classified into 7 land use/land cover classes in the case of optical and 10 classes in the case of SAR. The experimental results show a good performance of both methods in retrieving built-up and natural scenes. However, the advantage of the last method is mainly the facility of its operation since it does not need to set input parameters and the image retrieval is full automatic.