Classification of remotely sensed images in compressed domain

The amount of image data acquired by space-based remote sensing missions has increased phenomenally over the years. This poses severe constraints on storage and network bandwidth resources. Image compression methods are employed to overcome some of these problems. However, in order to perform any image processing operations (such as feature extraction, segmentation, spectral analysis etc.), images need to be decompressed first. Obviously, decoding or decompression requires more computational and storage resources. Also, this step does not produce new information. By directly operating on compressed images, we can eliminate the need for decompression and save time and space. In this paper, we present a framework to classify remotely sensed images in the compressed domain. Specifically, we propose a compression model based on Vector Quantization. Indices and codevectors that represent macro blocks of an image are exploited in the subsequent classification phase. Our experiments demonstrate that the proposed method is very efficient.