Feature detection via linear contrast techniques

摘要

A new technique for the detection of grey level features embedded in noise is developed using the linear contrast methodology within the context of the analysis of variance (ANOVA). The feature to be detected is first partitioned into a set of disjoint homogeneous regions. Based on the one-way model under ANOVA, each region is assigned a weight factor called an elementary contrast. A test statistic, which is simply a linear combination of the elementary contrasts, is generated and the feature detection problem is reduced to a hypothesis test with a confidence level α. The technique introduced here is particularly useful in real-time processing due to the computation of only one sum of squares in the alternative space. The problem of using additional test features is addressed by introducing orthogonal contrast functions. The procedure proved to be computationally efficient and very reliable in noisy environments.