A Critical Performance Evaluation of Classification Methods with Modified JPEG Decompressed Multiband Images
Effective utilization of bandwidth andstorage space is important in imaging applicationsincluding remote sensing. Remote sensingapplications use multi-sensory, multi-band, multiresolution images. Usually, remote sensingapplications uses image classification results fortheir analysis and decision making. In this paper wepropose a new JPEG based image compressionalgorithm based on filters. Proposed algorithmperformance is evaluated in relation to conventionalJPEG algorithm. In order to envisage the effect ofcompression on classification performance, MaximumLikelihood, Mahalanobis and Euclidean distanceclassifiers performance is evaluated with originalimage data, conventional JPEG compressed data andthe compressed image data with the proposedmethod. Experiments are carried out with manymulti-band images. Our experiments supports thatthe classification accuracies of compressed imagesare at par with original image data.
