Design of multi-layer perceptrons via joint filled function and genetic algorithm approach for video forensics

Abstract

This paper proposes a direct design of a multi-layer perceptron via a nonconvex optimization with application to the camera identification for the video forensics. Here, very low bit rate videos with time varying overall noise pattern statistics compressed using different video coders are considered. First, the content removal based on the discrete fractional Fourier transform is performed for video forensics. Then, the camera identification is performed via a multi-layer perceptron. To design the multi-layer perceptron, the design problem is formulated as a nonconvex optimization problem. Since the nonconvex optimization problem consists of more than one local minimum, finding the globally optimal solution of the nonconvex optimization problem is challenging. A joint filled function method and the genetic algorithm is developed for finding the solution of the nonconvex optimization problem.