Analysis of bilevel quantizers used in binary image correlators

摘要

Binary image correlators used for realtime scene matching demand reliable algorithms for quantizing the scenes being matched into meaningful binary images. The paper presents a mathematical analysis of two types of local average quantizer that are widely used in binary image correlator systems: line average quantizers and area average quantizers. An image is modelled as a two-dimensional discrete random field in which each random variable, assumed to have a Gaussian distribution, denotes the intensity of an image point. The probability of error, i.e. the probability of incorrectly quantizing a pixel, is used to evaluate the different quantizers. The analytical result reveals the strong and weak points of each quantizer and is helpful in determining the type of quantizer and the kernel size best suited for a given scene. Some simulation results are presented.