Adapting Interest Point Detection to Illumination Conditions
Flore Faille · 2003
Abstract. The objective of the presented work is to improve the stability of interest point detection under illumination changes. The use of a global threshold is shown to be insufficient and three methods are proposed to enhance a state–of–the–art algorithm: the Harris corner detector. These methods are based on different principles: a local image normalization as preprocessing, a local threshold adaption, and a local automated threshold selection based on clustering. All methods are compared on several image series created by varying the lighting conditions. For performance evaluation repeatability and false positive rates are used. All methods allow a stability enhancement under complex illumination changes. The algorithm based on threshold adaption performs best. 1