Atalar M., BİLGE H. Ş.

JOURNAL OF THE FACULTY OF ENGINEERING AND ARCHITECTURE OF GAZI UNIVERSITY, vol.24, no.1, pp.171-181, 2009 (SCI-Expanded) identifier identifier


Motion estimation is usually performed in the spatial domain at the studies in the literature. In this study, motion estimation in frequency domain by using pseudo phase calculation is investigated. Pseudo phase calculation is defined as the extraction of the phase information by using coefficients belonging to frequency domain (Discrete Cosine Transform and Discrete Sine Transform) which don't really include any phase information. The method of motion estimation by using pseudo phase is successful in clean and synthetic images. In this study, the unsuccessful conditions are analyzed and some enhancements are proposed in order to have success even in these conditions. In experimental investigations, it is seen that the calculation of the pseudo phase is negatively affected by the noise on images. To avoid this drawback, it is proposed that the pseudo phase is calculated after the noise reduction by a median filter and hence more successful results are obtained. When the gray level difference between the object and its background is low, then the result of the method is also wrongly affected. After contrast enhancement (e. g., exponential transformation), it is seen that the success of the pseudo phase calculation is increased. Furthermore, after adaptive thresholding (e. g. Otsu's method) successful results are again obtained. As a result, robust motion estimation is successfully applied by pseudo phase calculation even under severe conditions.