The goal of a machine perception system is to automatically interpret the information in physically sensed signals, such as sound, light, radar, or pressure signals, in a useful way for a given application. Researchers have attempted with limited success to use digitized signals from various sensors to endow digital computers with the abilities to hear, see, and touch so that machines can intelligently interact with an uncertain, dynamic environment. Whereas automated speech recognition techniques attempt to understand one-dimensional (1-D) signals in terms of a natural language structure, computational vision research attempts to use two-dimensional (2-D) signals to determine the geometric structure of corresponding three-dimensional (3-D) scenes. Each data point in a digitized 2-D signal is discrete in the two x, y spatial directions and in the level z of the sensed quantity. Such 2-D signals are usually processed as a large matrix of integers. Each matrix element, or pixel, has a row/column location and a value representing the sensed physical quantity. In machine perception, it is useful to view a large matrix of this type as a digital surface because the sensed values at each pixel can be considered as noisy samples of an “analog surface” z = f(x, y). Digital surfaces are more commonly known as digital images. Figure 1.1 shows a 20×20 matrix taken from a larger digital image and the corresponding digital surface view of the same data.
KeywordsImage Segmentation Intensity Image Range Image Surface Patch Image Segmentation Algorithm
Unable to display preview. Download preview PDF.