A shot is a continious sequence of frames filmed in one go. The regions of discrepency between two shots is called the shot boundary.
Various discriminatory features like the frame variance, mean, L2-norm etc are taken and then various classifiers are used to compare their performances.
Best performance is achieved by The Linear SVM with a classification accuracy of 92.99%.
A small library for feature detection from images, along with the best features chosen to do shot boundary detection is included. If used, please give credit to me. If you want to use the code you'll be on your own in understanding and assimilating into your own work.