The image node: An image processing neural network component

A specialized variation on the artificial neural node is proposed that can accept unprocessed image data at its inputs. The intended use of the neural node variation, called an Image Node, is to efficiently replace a large number of conventional neural nodes at the lowest level of artificial neural networks used in image recognition devices. Prior applications of artificial neural networks have required special preprocessing and the transformation of image data before the data could be presented to the network. The Image Node incorporates conventional image processing algorithms that permit it to accept unprocessed color image data and to make comparisons with a stored exemplar.; The Image Node based network bears some resemblance to a Counter-Propagation Network, which consists of a lower layer of LVQ neurons and an upper layer of Outstar neurons. The Images Nodes have a role that is similar to the LVQ neurons—learning a set of codes, or in this case, image fragments, which can encode the target image set. In place of the Outstar layer of a Counter-Propagation Network, a network based upon Image Nodes will use one or more upper layers of back-propagation nodes.; The image Node has internal mechanisms to deal with problems of color comparison and background versus foreground image information. It's design also emphasizes the importance of localized areas of image match. These heuristic functions give each Image Node the power to replace a very large number of conventional neural nodes. The Image Node is proposed as a potentially useful tool to be applied to image recognition tasks. The Image Node algorithm is described and demonstrations of its capabilities are presented.