Automated landmark extraction in digital images - performance evaluation

In this study, we present an automated system for feature recognition in digital images. Such a system is very important for morphometric analysis as it will replace the time consuming and labour intensive process of manual identification. The analysis system is constructed from four key stages: a feature based detection of the fly wing structure, recording the compact invariant shape descriptors using the pairwise geometric histogram (PGH) representation, global estimation of the pose using the probabilistic Hough transform and finally a correlation based refinement of individual features. The performance of the system, its reliability and robustness are evaluated in comparison to the expert manual performance.