Relocated I-Frames Detection in H.264 Double Compressed Videos Based on Genetic-CNN
Qiang Xu, Xinghao Jiang, Tanfeng Sun, Peisong He, Shilin Wang, Bin Li · 2018
Analyzing the appearance of Relocated I-frames is a vital step in double compression detection in Group of Pictures (GOP) non-aligned videos. In this work, a frame-wise relocated I-frames detection method in H.264 double compressed videos based on Genetic-CNN is proposed. Video clips which contain three adjacent frames are used as the input of network to separate image content from noise. A preprocessing operation is adopted by extracting the noise residual. The genetic algorithm is applied to verify the possibility of automatically designing deep network structures. The network optimization operation mainly includes CNN encoding, initialization, the construction of fitness function and genetic operations, e.g. selection, mutation and crossover. By testing on a data set composed of published YUV sequences, the results clearly demonstrate the efficacy of the proposed approach and show that the generated CNN can achieve better performance than previous method investigated.