Nonlinear filters in image pyramid generation

The possibility of using nonlinear edge- and detail-preserving median-type filters for image pyramid generation is investigated. Nonlinear interpolation filters should make better use of image structure and the differences between original and interpolated images can be expected to have low entropy as the image information will be removed from them to a high extent. However, there are many other factors which have impact on the entropy of the difference images and the theoretical expectations must be checked experimentally. Several types of detail-preserving median-type filters for interpolation are described. These filters are used in various combinations for the generation of image pyramids. Entropy of differential images is measured and compared to linear filters. The results show that the nonlinear filters perform well on various types of images and give results which are close to the best linear filters.<>