2000 年 66 巻 2 号 p. 261-265
A robust approach to the half-tone template-matching problem is proposed. It is based on a novel statistic called 'Increment Sign Correlation (ISC)' which is also proposed in this paper. ISC is formulated mathematically as a statistic which is a kind of cross correlation between binary coded increment signs and has a probabilistic binary or normal distribution. Characteristics of the distribution depend on both average contrast of the template picture and variance of additive noise. The approach utilizes ISC effectively for matching partial pictures with the template. It is applied to recognition of glossy objects with highlights and detection of sudden change in natural scene. Robustness of matching methods is much necessary in these tasks for avoiding brightness errors almost only by illumination. Experiments with real pictures show an effectiveness of the proposed approach.