Search

In this chapter an overview of models for motion detection and contrast enhancement in early vision is presented. Although many other visual tasks can be studied for VLSI implementation, most of the issues are similar to those of these two specific tasks that are addressed here, i.e. motion detection and contrast enhancement, and many of the circuits and building blocks can be reused.

Hypermedia systems have become increasingly popular however, the early vision of hypermedia [4] as a panacea for easy access to large quantities of information has proven elusive. Many users complain of being “lost in hyperspace” and most current hypermedia systems must supplement their basic navigation strategies with search mechanisms (see Chapter 2 of this book). Even so, looking for information in a hypermedia system can be frustrating. Traditional interaction styles provide users with limited views of the information space being explored and hide most of the information (and its organisation). Users often have no idea of what information is contained within a web of documents until they actually find it. This leads to both navigation errors and badly formulated queries.

A good deal of recent research into sensory information processing algorithms has focussed on the so-called energy function minimization, or regularization, approach to inverting the world-image mapping. In this chapter we review the application of this paradigm to early vision and its incorporation within the Bayesian framework. We will concentrate in this section on using this formulation to impose smoothness constraints on the solutions; other constraints can be similarly imposed, as will be described in later sections. A large part of this chapter will be concerned with new approaches to formulating energy function based algorithms using techniques borrowed from statistical mechanics.

Convolution, one of the most widely used functions in early vision processing, is, at the same time, highly demanding in terms of computing power. This paper describes a dedicated coprocessor able to perform ID and 2D convolutions; the latter case is applicable only if the convolving bidimension- al filter is separable. Details concerning the main building blocks are given together with experimental results showing the overall performances of the system.