This paper introduces an approach for the extraction and combination of different cues in a level set based image segmentation framework. Apart from the image grey value or colour, we suggest to add its spatial and temporal variations, which may provide important further characteristics. It often turns out that the combination of colour, texture, and motion permits to distinguish object regions that cannot be separated by one cue alone. We propose a two-step approach. In the first stage, the input features are extracted and enhanced by applying coupled nonlinear diffusion. This ensures coherence between the channels and deals with outliers. We use a nonlinear diffusion technique, closely related to total variation flow, but being strictly edge enhancing. The resulting features are then employed for a vector-valued front propagation based on level sets and statistical region models that approximate the distributions of each feature. The application of this approach to two-phase segmentation is followed by an extension to the tracking of multiple objects in image sequences.