Image compression method for preprocessing image based on learning frequency domain information

Abstract

The invention belongs to the technical field of image processing and neural networks, and particularly relates to an image compression method for preprocessing an image based on learning frequency domain information. Through the deep learning technology, the image needing to be compressed is pre-processed, then the pre-processed image is compressed, and the effect of the corresponding compression method can be directly improved on the premise that the compressed image does not need to be further post-processed. According to the method, the frequency domain information of the input image is learned through the neural network, and the corresponding relation between the frequency domain information and the compressibility of the image is established, so that the neural network can balance between the compressibility and the quality loss of the image, and the input image can be properly preprocessed to improve the compression effect.

Description

technical field [0001] The invention belongs to the technical field of image processing and neural network, and in particular relates to an image compression method for preprocessing images based on learning frequency domain information. Background technique [0002] The principle of pre-processing / post-processing optimization for image compression: treat the image encoder as a black box, pre-process the image to be compressed before compression to obtain a pre-processed image, use the pre-processed image for compression, and decode and reconstruct the image after decoding. Post-processing operations are performed on the reconstructed image to improve the quality of the reconstructed image. [0003] Therefore, the preprocessing operation is used to reduce the complexity of the image to be compressed, and it can use fewer code words (BPP (bitsper pixel, the average number of bits per pixel), which indicates how many bits per pixel need to be used on average after encoding an ...

AI Technical Summary

Problems solved by technology

[0011] "Klopp J P, Liu K C, Chen L G, et al. How to Exploit the Transferability of Learned Image Compression to Conventional Codecs[C] / / Proceedings of the IEEE / CVF Conference on Computer Vision and Pattern Recognition. 2021:16165-16174." Adopted The codec based on deep learning approximates the rest of the codecs, simulating the BPP information of the input image after its compression, "Talebi H, Kelly D, Luo X, et al. Better compression with deep pre-editing[J]. IEEE Transactions on Image Processing, 2021, 30: 6673-6685." directly uses a differentiable JPEG image codec to help the preprocessing network to train. However, these two methods are only approximate alternative methods, only It can approximate the BPP size after image compression, and its estimated value does not have a high enough correlation with the real BPP value after image compression

In addition, the complexity of these two loss functions is relatively high, which greatly increases the complexity of network training.

Images