Colour Histogram and Modified Multi-layer Perceptron Neural Network based Video Shot Boundary Detection

DaltonThounaojam¹, Thongam Khelchandra², Thokchom Jayshree², Sudipta Roy³, and Khumanthem Singh²

¹Department of Computer Science and Engineering, National Institute of Technology Silchar, India

²Department of Computer Science and Engineering, National Institute of Technology Manipur, India

³Department of Computer Science and Engineering, Assam University Silchar, India        

Abstract: The paper proposes a shot boundary detection technique using colour histogram difference and modified Multi-Layer Perceptron (MLP). In this the learning process in the MLP is modified as an evolutionary learning process using Genetic Algorithm (GA) in which the weights of the hidden layer and output layer of the MLP are updated by GA. Colour Histogram Differences (HD) between two consecutive frames are used for feature extraction. Four values HD_i,HD_i-1 and-1 are used as an input for the modified MLP Neural Network where HD_i is the colour histogram difference between frame f_i and f_i+1, HD_i-1 is the colour histogram difference between frame f_i-1 and f_i and HD_i+1 is the colour histogram difference between frame f_i+1 and f_i+2. The propose system is tested with the TRECVid 2001 and 2007 test data and it is also compared with latest algorithms and yields better results.

Keywords: Abrupt; fade-in; fade-out; dissolve; shot boundary detection; neural network; genetic algorithm.