Using ICA as a Pre-Processor for CNNs to Efficiently Process High-Dynamic Videos

Convolutional neural networks (CNNs) require long-term iterations to get the optimal solution, where convolution operation consumes a lot of time. In this work, Independent Component Analysis (ICA), which is a widely used statistical analysis tool, is used to pre-process the video to present the input data for CNNs. ICA is employed to take the place of one or more convolutive layers in CNNs. By extracting multi-channel component representing different aspects of the raw images, ICA provides CNNs with processed images which contain rich and apparent features. Quantitative experimental results indicate that the use of ICA allows the same or even better performance to be achieved in less time.