The effect of compression techniques on classification accuracy

Satellite image and air photograph are basics of remote perception. Also, these are main sources of location based projects. Nowadays, digital systems substitute for analog and analytic systems with developing of computer technology. However, image processing, transfer and storage started to be major issue. At the end of this study, Landsat 8 (10 band, 30m resolution) satellite images were used to get new images. New images being compressed lossless were obtained in MrSID, ECW and JPEG2000 formats. Also, other new images were obtained as being compressed with ratios of 10:1, 20:1, 30:1, 40:1, 50:1/60:1 as being lossy. The original and compressed images were classified base on pixel (the greatest classification method). In order to detect the accuracy of classification; data obtained by classification must be compared with the reference data that represents real surface images. Ortophoto images that were obtained by air photographs which were taken by Vexcel UltracamX digital air camera (4 band, 30 cm resolution) were used as reference data. At 120 reference points, lossless compressed reference images were compared with reference lossy compressed images in different ratios to detect classification accuracy. At the end of these studies, especially images which were compressed lossless and in low ratios can be used for classification.