Robust region merging for motion based segmentation using the Kolmogorov-Smirnov test

• Autorzy: Broekhoven M. , Hupkens T. , Hendriks E. , Patras I.
• Języki publikacji: EN

Abstrakty

EN

This paper addresses the problem of segmenting image sequences into moving objects. The proposed method uses a bottom-up approach and uses an indirect method to estimate the motion parameters of the initial regions. These regions have been created using the watershed algorithm. The motion of the regions have been modelled by a four parameter affine model and have been estimated by using a robust method, which diminishes the influence of wrongly estimated displacement or optical flow vectors (outliers). The similarity measure between two adjacent regions is computed through a non-parametric statistical test (the Kolmogorov-Smirnov test) as proposed by Moscheni [1] with some important modifications. The algorithm iterates between a motion estimation step and a region merging step until some stopping criterion has been reached. The results indicate that the method is well suited to obtain a correct segmentation of real image sequences.

Słowa kluczowe

EN

Kolmogorov-Smirnov test; motion estimation; pattern recognition; region-based segmentation

PL

estymacja ruchu; rozpoznawanie obrazów; test Kolmogorova-Smirnova

• Wydawca: Instytut Telekomunikacji i Informatyki Uniwersytetu Technologiczno-Przyrodniczego w Bydgoszczy
• Czasopismo: Image Processing & Communications
• Rocznik: 2000
• Tom: Vol. 6, no 3-4
• Strony: 27--34
• Opis fizyczny: Bibliogr. 14 poz.,Rys., wykr.,

Twórcy

autor

Broekhoven M.

autor

Hupkens T.

autor

Hendriks E.

autor

Patras I.