Classification on compressed images with bounded loss

The influence of lossy compression is evaluated by using a classification algorithm on high resolution two-band images of vegetation canopies. The experiments were aimed at evaluating the feasibility of using lossy compression techniques to save on the immense storage requirements of remotely-sensed images. The images were compressed using a discrete cosine transform based approach, at various degrees of loss and compression. An unsupervised classification was performed with ERDAS to stratify each image into ten classes. The classification results are compared with those obtained from the original image without compression. This comparison is done by plotting the percentage of pixels classified differently as a result of the loss.