Use of support vector machine based on color and motion features for shot boundary detection

C.Dalatsi, S.Krinidis, S.Tsekeridou, I.Pitas

International Symposium on Telecommunications, IST01, Iran.

Abstract: In this paper we present a shot cut detection technique based on the fusion of information provided by three individual methods of analyzing both the static and the dynamic features of video se­ quences. The presented video analysis focuses on extracting mul­ tiple low­level content descriptors that once combined, enhance the shot boundary detection results achieved when each descriptor is considered individually. Static feature analysis provides the sys­ tem with the bin­wise differences of color vector histograms and also with the color frame differences, both computed between suc­ cessive frames. Furthermore, dynamic feature analysis yields bin­ wise differences of motion vector histograms calculated on suc­ cessive frames too. Static and dynamic features are combined to form 4D feature vectors characterizing a frame in the temporal dimension. Shot cuts are detected via the implementation of a su­ pervised Support Vector Machine classifier so that feature vectors are assigned to two classes, one representing shot cuts and another representing in­shot information. The method proves to behave satisfactorily leading to remarkable recall and precision values.