Morphological filters implementation based on a co-design approach

In the framework of object-based video coding morphological filtering has became an important technique for image simplification. Morphological filters are used in a pre-processing step to improve the chances of obtaining meaningful segmentation results. However the computation cost of these filters limits their practical use. An improvement of the processing time is then required for their use in real-time applications. In this paper the implementation issue is addressed. Improvement of the processing time is achieved by exploiting the local parallelism of these filters and by reducing the number of clock cycles used to access a specific data structure called the hierarchical queue. As various morphological filters can be used, an architecture based on hardware software partitioning is proposed. Performance results showing the benefits of the proposed architecture are provided.