Multiresolution lossless image compresson using the predictive S transform

Abstract

The sequential transform(S Transform) is simple and significantly reduce the first order entropy. The transform leaves a residual correlation between high pass components which is due to aliasing from the low frequency components of the original image. This aliased components are reduced by using better filters. Also, the residual correlation can be reduced by using the prediction. In the thesis, new predictive sequential transform is proposed, which is using interpolation and prediction. By the interpolation and difference estimation, new predicted difference values are obtained. From this predicted value, the difference between original and the predicted ones are calculated. This differential image typically has much reduced variance and can be encoded more efficiently.