KYDON - AN AUTONOMOUS, MULTILAYER IMAGE-UNDERSTANDING SYSTEM - LOWER LAYERS

In this paper, an autonomous, application-specific system architecture , called KYDON, is presented. The KYDON architecture consists of k lay ers of array processors. The lowest four layers compose the KYDON's lo w-level image-processing group, and the rest of the layers compose the higher-level processing groups. The interconnectivity at each array p rocessor is based on a full hexagonal mesh structure. Each processing element (PE) of an array processor is a simple autonomous unit, includ ing its own control unit (CU). This paper deals with the internal stru ctural design of the PEs at the lower layers. It also provides a descr iption of the low-level image-processing tasks performed by KYDON's lo wer-array processors. More specifically, the lowest layer is a 2-D pho toarray, which captures images from the environment. The next three la yers perform image processing and generate the graphic forms of the ob jects extracted from the input image. The next layers process the grap hs provided by the lower layers in order to achieve an image understan ding of the input image. An important feature of KYDON is that KYDON d oes not have any host computer or control processor to handle I/O and other miscellaneous tasks.