In this chapter we show the results of our comparison of the linear, the quadratic, the convolutional neural networ, and the back-propagation neural network and demonstrate a method to extract powerful subsets of the features used to describe still images and video frames. Good feature sets can be found that preserve much of the robustness of the entire feature set using only about a quarter of all features. Sections 3 and 4 in this chapter show the performance of the various intermediate components of the framework as well as the final classification and event detection results.
KeywordsGreedy Algorithm Neural Network Classifier Shot Boundary Shot Boundary Detection Rocket Launch
Unable to display preview. Download preview PDF.