The aim of Windowing is to take image or volume data with a high number of brightness values and to reduce these to just 256 brightness values, so that they may be displayed on computer screens or printed. We present an algorithm which is based on the well known Reinhard method and extend it to work on a large variety of data. In our focus are normal images, medical volumetric datasets and curvilinear grids. We compare our algorithm with other existing methods from Ashikhmin and Durand to demonstrate its quality and artefact freeness.

The algorithm works in both two and three dimensions. Because the speed of gaussian filtering (the central part of the algorithm) is especially crucial with large medical volumes, we tested different approaches to maximise its speed. For 3D volumes a separable convolution in the spatial domain proved to be superior to a frequency domain convolution with fast fourier transformations. Our implemented algorithm works in linear time (not cubic) for any width of the 3D filter kernel.

Report  

Full featured description of the program and the algorithms in the PDF format. Optimized for double-sided printing. Update 2004/12/05. (German)