Classical video prediction methods exploit directly and shallowly the intra-frame, inter-frame and multi-view similarities within the video sequences;the proposed video prediction methods indirectly and intensively tr...
详细信息
Classical video prediction methods exploit directly and shallowly the intra-frame, inter-frame and multi-view similarities within the video sequences;the proposed video prediction methods indirectly and intensively transform the frame correlations into nonlinear mappings by using a general deep neural network (DNN) with single output node. Traditional DNN based video prediction algorithms wholly and coarsely forecast the next frame, but the proposed video prediction algorithms severally and precisely anticipate single pixel of future frame in order to achieve high prediction accuracy and low computation cost. First of all, general DNN based prediction algorithms for intra-frame coding, inter-frame coding and multi-view coding are presented respectively. Then, general DNN based prediction algorithm for unified video coding is raised, which relies on the preceding three prediction algorithms. It is evaluated by simulation experiments that the proposed methods hold better performance than state of the art High Efficiency videocoding (HEVC) in peak signal to noise ratio (PSNR) and bit per pixel (BPP) in the situation of low bitrate transmission. It is also verified by experimental results that the proposed general DNN architecture possesses higher prediction accuracy and lower computation load than those of conventional DNN architectures. It is further testified by experimental results that the proposed methods are very suitable for multi-view videos with small correlations and big disparities. (C) 2017 Elsevier B.V. All rights reserved.
作者：
Li, HongguiYangzhou Univ
Sch Informat Engn 196 West Huayang Rd Yangzhou Jiangsu Peoples R China
This study proposes a framework of videocoding based on Laplacian eigenmaps (LEM) and its related embedding and reconstruction algorithm (ERA). Firstly, a one-dimensional (1D) representation of LEM is adopted to achi...
详细信息
This study proposes a framework of videocoding based on Laplacian eigenmaps (LEM) and its related embedding and reconstruction algorithm (ERA). Firstly, a one-dimensional (1D) representation of LEM is adopted to achieve an extremely low bit per pixel (BPP). Secondly, dualk-nearest neighbours, which keeps neighbour relationships both in high-dimensional data space and low-dimensional representation space and overcomes the disadvantage of classical non-linear dimensionality reduction methods which cannot preserve the neighbour properties in both of the spaces, based ERA of LEM is employed to gain extraordinarily high peak-signal-to-noise ratio (PSNR). Thirdly, a unified framework of videocoding is fit for intra-frame, inter-frame and multi-view videocoding. Finally, it is evaluated by simulation experiments that, in the situation of low bitrate transmission, the proposed method can attain better performance of BPP and PSNR than that of the state-of-the-art methods, such as highly efficient videocoding.
暂无评论