Video Shot Boundary Detection Based on Localized Multiple Kernel Support Vector Machine

Video shot boundary detection approach based on localized multiple kernel support vector machine(L-MKSVM) is proposed.Intermediate features are constructed by integrating local temporal-spatial information of video frames,and then video frames are split into boundary and non-boundary frames with L-MKSVM.In order to improve the detection precision of MKSVM based on global optimization,video frames are projected to Hashing subspace with locality sensitive Hashing(LSH),and then L-MKSVM is constructed for each Hashing subspace with multiple kernel learning methodology.The synthetic minority over-sampling technique(SMOTE) is used to solve the imbalance problem of video boundary frames and non-boundary frames.The experimental results show that the introduced shot boundary detection approach can improve the precision and recall.