A Perceptual Image Compression Method Based on Region-level JND Prediction

Abstract

The present invention relates to a perceptual image compression method based on regional block-level JND prediction, comprising the following steps: 1) generating regional block-level JND values ​​according to images in a data set and corresponding JND information using the Otsu threshold method; 2) generating 3) Compress the test image under multiple fixed QF values ​​to obtain multiple corresponding distorted images, and divide all the distorted images into multiple Non-overlapping area blocks, and predict the JND label of each area block, and finally use the label processing method to obtain the final JND value of each area block; 4) According to the target compressed QF value and the final JND value of each area block, test The image is preprocessed, and the largest perceived QF value of the regional block is selected as the compression parameter, and the preprocessed test image is compressed by JPEG. Compared with the prior art, the present invention has the advantages of self-adaptive prediction, good compression quality, high compression efficiency and the like.

Description

technical field [0001] The invention relates to the field of image compression, in particular to a perceptual image compression method based on region block-level JND prediction. Background technique [0002] With the development of social network and multimedia technology, a large amount of picture information is generated on the Internet. According to recent statistics, Instagram users upload about 90 million pictures every day. Therefore, how to store and transmit these images is an extremely challenging task. Existing image compression standards, such as JPEG, H.264 and HEVC, all use PSNR and MSE as the standard to measure distortion. In the process of calculation, it is considered that each pixel is equally important, which is inconsistent with the human visual system. Therefore, it is particularly important to study the image compression algorithm for the human visual system. [0003] Various methods have been proposed to solve this difficulty, including JND-based me...

AI Technical Summary

Problems solved by technology

Although the existing perceptual coding model can reduce the perceptual redundant information in coding to a certain extent, it only considers limited visual characteristics and does not change with the change of quantization parameters. The latest perceptual experiments show that the human visual system The perception of quality presents a ladder shape and does not change continuously. Each mutation point can be regarded as a JND value

However, for an image, a large number of subjective experiments are required to obtain the final JND value, which cannot be applied in reality. In addition, different image contents should contain different JND values. Therefore, the design of block-level JND prediction and compression algorithms seems particularly necessary

Images