Effective utilization of bandwidth and storage space is important in imaging applications including remote sensing. Remote sensing applications use multi-sensory, multi-band, multi resolution images. Usually, remote sensing applications uses image classification results for their analysis and decision making. In this paper we propose a new JPEG based image compression algorithm based on zooming-shrinking technique. Proposed algorithm performance is evaluated in relation to standard JPEG algorithm. In order to envisage the effect of compression on classification performance, Maximum Likelihood, Mahalanobis and Minimum distance classifiers performance is evaluated with original image data, standard JPEG compressed data and the compressed image data with the proposed method. Experiments are carried out with multi-band images with various resolutions. Our experiments supports that the classification accuracies of compressed images are at par with original image data.