Adding explicit content classification to nonlinear filters

Nonlinear filters are known for better edge-preserving performance in image processing applications as they can adapt to some local image content. Instead of trying to find a single optimal filter that can adapt to all the image content, some classification-based approaches first apply a pre-classification on the image content and then employ an optimal linear filter for each content class. It is interesting to extend the linear filter in such approaches to a nonlinear filter and see if the explicit content classification, can still add to such inherently adapting nonlinear filters. In this paper, we investigate several categories of nonlinear filters: order statistics filters, hybrid filters, neural filters, and bilateral filters with different forms of content classification in various image processing applications, including image de-blocking, noise reduction, and image interpolation.