Research On Video Intra Coding Technology Based On Preprocessing Optimization

Lossy compression that trades off distortion and bit rate budget is the mainstream choice for compressing image and video content today.For the two trade-offs of distortion and bit rate in lossy compression,the latter is objective,while the former has not been uniformly defined.From the perspective of human vision,there is still no completely objective distortion measure to reflect the distortion of human vision.In addition,with the rapid development of computer vision technology based on deep learning in recent years,a large number of machine analysis systems have been implemented.Video and image content not only serve the needs of human vision,but also need to serve the needs of machine vision.The definition of machine vision distortion in different scenes is also inconsistent,so the encoder should also consider different distortion characteristics.Because the design of the traditional coding method is relatively complicated,modification under different distortions will cause changes at the encoder end and the decoder at the same time,and the modification cost is high.Therefore,it is a reasonable alternative to process the compressed content according to different distortion characteristics through pre-processing technology before encoding,and improve its performance under different distortions without changing the subsequent encoding method.In this thesis,the HEVC coding method in intra-frame mode is used as the compression method to be optimized,and the convolutional neural network is used to build a preprocessing model.Referring to the characteristics of the discrete cosine transform used by the HEVC encoder,the frequency coefficients after the discrete transformation of the image are used as the network input,and the channel attention mechanism of the neural network is used to learn the preferences of the distortion measure on the different frequency information of the image under the definition of different distortion loss functions.According to this,the content of the image to be encoded is processed to suppress the relatively unimportant frequency information,so that the HEVC encoding method can use the same bit rate budget to save the frequency information preferred by the distortion,thereby improving the performance of the HEVC encoding method under different distortion definitions.In this thesis,the performance of the proposed preprocessing method is fully evaluated and compared under the definition of distortion in different scenarios,and the the effectiveness of the proposed method is shown by the results of this thesis's experiments.In addition to the HEVC encoding method in intra-frame mode,which is mainly used for comparison,this thesis also evaluates the improvement performance of the proposed method on other traditional image compression methods.Experimental results show that the performance of traditional image compression/video intra-coding methods can be flexibly improved by reasonably preprocessing the image content under different distortion definitions.