Image partition generation with region boundary encoding

We introduce an image segmentation system designed to model images based upon aspects of the human visual system. The initial partition is generated by a Watershed Transform (WST). The final partition is determined after applying region reduction techniques. All aspects of the image segmentation system are automatic. We show the efficacy of the segmentation system and subsequent region boundary encoding by using the resulting partition in a region-based image compression algorithm. Our results show that the image coding efficiency based on our segmented model is near that of other lossless techniques despite the added requirement of encoding region boundaries.; The production and encoding of the image partition is divided into several distinct processing steps. At each processing step, we describe new methods to facilitate the automatic generation and encoding of the final partition. We provide analysis regarding the uniqueness of the WST-generated partition. We introduce marklet coupling filters and drainage simulations as automatic methods to mitigate WST oversegmentation. We present a unique multiphase region-merge algorithm that eliminates the need for constant monitoring of merge termination criteria. We introduce an efficient method to generate and losslessly encode region boundaries. Our pixel-based chain coder introduces the novel techniques of dynamic template switching and extended vector templates. We show how region-based parameters can be used to improve decompressed image quality on a per-region basis.