TRACE TRANSFORM AND KLT BASED INVARIANT FEATURES AND IMAGE RECOGNITION SYSTEM
Ján Turán, Peter Filo, B. Němcovej, Kaiser Passage · 2006
An important problem in invariant image recognition is automatic extraction of invariant features, which are able to describe an object invariant under translation, rotation and scale. Elegant solution of this problem is using Trace transform based on Triple Features. To compute invariant Triple Features a sophisticated combination of three functionals (Trace, Circus and Diametric) is needed. The paper overline the properties of the functionals must have in order to extract invariant features. The problem of high dimensionality and selecting of optimal combinations of functionals is solved using Karhunen – Loeve transform (KLT). The new architecture of invariant image recognition system based on Trace transform and KLT is presented. The system performance was tested in image recognition experiments.