Image Processing Techniques: Some Possible Applications to the Transduction of Visual Information to Other Modalities

Image processing techniques may be classified in terms of a rather loose notion of complexity. Firstly, there are bulk transformations where the one image is translated into another by applying the same rule to each pixel in the source image. Examples of this type of process are differentiation and spatial summation. This type of process is amenable to parallel computation. At a slightly higher level is feature detection: the selection of points satisfying some criteria. Examples of this type of process are the search for and classification of edges, ends of lines, corners and forks. At the next level evaluation may concern compound structures: the spatial relation of simple features to form a new identity. The relations that may be used are to the right of, above, next to, enclosed by etc. For example a square is an edge which is closed, with four corners, and each corner is approximately 90 degrees: a tea cup in profile might be defined by a square next to a circle. Another issue is the handling of the amount of information involved in image processing. An image potentially contains a large amount of information. A picture 512 by 512 pixels with 8 bit colour definition in red, green, and blue will take 0.75 Megabytes of storage. This is roughly equivalent to the storage needed for a book. The human eye scans a picture concentrating on the dominant features and spending variable amounts of time extracting information from the various regions of a scene.