A Tensor Based Unified Framework for Streaming Signal Online Analysis

To explore the online tensor analysis technique in single-aspect streaming model, a tensor based unified framework encompassed more tensor formats than one is proposed according to the structure of the last tensor factor. We found that the temporal slice only changes the size of factor matrix corresponding to the temporal dimension in unified framework. Consequently, the block-coordinate descent strategy is utilized to calculate the temporal tensor factor and non-temporal decomposition components respectively based on exponentially weighted cost function. The simulation results indicate that the presented unified framework provides a superior reconstruction performance compared to the existing tensor tracking methods in real video scenarios.