Method for constructing convolutional neural network for video coding fractional pixel interpolation

Abstract

The invention provides a method for constructing a convolutional neural network for video coding fractional pixel interpolation, which comprises the following steps: images with different content andresolution are collected, and an original training data set containing data with different types and coding complexity is formed; preprocessing operation is performed on the original training data setto obtain training data conforming to the video coding inter-frame prediction fractional pixel interpolation characteristic; a deep convolutional neural network is built to obtain a convolutional neural network structure suitable for the video coding inter-frame prediction fractional pixel interpolation; the pre-processed data is input into a built-up convolutional neural network; meanwhile, theoriginal training data set is used as a corresponding true value to train the built-up convolutional neural network. According to the method, the convolutional neural network can be successfully trained; the fractional pixels obtained by using the trained convolutional neural network interpolation meet the requirement for video coding fractional pixel interpolation characteristic; and the method in the invention is used for performing fractional pixel interpolation so that the video coding efficiency can be improved.

Description

technical field [0001] The invention relates to a method in the technical field of image processing, in particular to a convolutional neural network method suitable for inter-frame prediction fractional pixel interpolation of video coding. Background technique [0002] Inter-frame prediction is a key technology in video coding standards. By using the similarity of video content between frames, the temporal redundancy of video can be effectively removed, thereby improving the coding and compression efficiency. At the same time, due to the discrete sampling operation in the digitization process, the real object motion does not necessarily follow the sampling grid. In order to further improve the accuracy of object motion prediction, in video coding standards, the motion of objects is in units of fractional pixels. The pixel values ​​at the fractional pixel positions on the sampling grid do not really exist. In the application, the pixel values ​​at these fractional pixel posi...

AI Technical Summary

Problems solved by technology

In addition, for the inter-frame predicted fractional pixel interpolation, the pixels at the fractional position do not really exist. Therefore, during the training process of the convolutional neural network, there is no real truth value to refer to, resulting in the normal training.

Images