Fondazione Bruno Kessler - Technologies of Vision

contains material from
Template Matching Techniques in Computer Vision: Theory and Practice
Roberto Brunelli 2009 John Wiley & Sons, Ltd

Chapter 4
Robust Similarity Estimators

A major issue in template matching is the stability of similarity scores with respect to noise, including unmodeled phenomena. Many commonly used estimators suffer from a lack of robustness: small perturbations in the data can drive them towards uninformative values. This chapter addresses the concept of estimator robustness in a technical way presenting applications of robust statistics to the problem of pattern matching. The approach is mainly based on the concept of influence function. M-estimators are discussed and L1 based robust correlation measures introduced. A solution to the problem of robust estimation of covariance matrices is discussed.

keywords: robustness, influence function, M-estimators, breakdown point, robust correlation coefficient.

 4.1 Validity of distributional hypotheses
 4.2 Tanh estimators
 4.3 L1 similarity measures