Abstract
This paper puts forward a new approach to second generation image coding. The concept of "image feature" is introduced in order to deal with those "objects" that cannot be properly described as regions within a segmentation framework. Visual features such as open contours or texture details are extracted from the original images using morphological operators. As malhematical morphology deals with the shapes and structures of the images, the resulting features are very close to what would be obtained from a visual perception point of view. If these features are efficiently coded with suitable techniques, such coding model is able to reach higher compression ratios than purely segmentationbased techniques. Numerical results are given at the end of the paper to prove the goodness of the feature-based coding scheme in a very-low bit-rate video coding application.