Frequency regularization reversible gray scale technology based on prior knowledge

CN119228939AInactive Publication Date: 2024-12-31QILU UNIVERSITY OF TECHNOLOGY (SHANDONG ACADEMY OF SCIENCES)
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202411769635.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-12-04
Publication Date
2024-12-31
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The existing reversible grayscale technology is difficult to effectively hide high-frequency information, which limits the quality of color image reconstruction, and the visual effect of the reconstructed color image is poor.

Method used

Introduce a high-frequency information extraction module, combine high-frequency information with the generated grayscale image as input, and train end-to-end through the color decoding and shading network of the UNet framework, using cross-entropy loss, brightness, contrast, local structure loss and quantization Loss function optimizes the network, forcing the decoder to fuse the high-frequency mode and improves reconstruction capabilities.

Benefits of technology

The visual quality of the reconstructed color images is improved, the SSIM and PSNR indicators are improved, and the visual quality of the generated grayscale images and the reconstruction effect of the color images is ensured.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119228939A_ABST
    Figure CN119228939A_ABST
Patent Text Reader

Abstract

The invention discloses a frequency regularization reversible gray level technology based on prior knowledge, and the technology is characterized in that the technology comprises the following steps: S1, taking an original color image as the input of a UNet framework-based decoloring network, and obtaining a generated gray level image; s2, a high-frequency information extraction module is introduced, gray level conversion and gray level conversion after Gaussian blur are carried out on the original color image respectively, and high-frequency information is obtained; s3, taking the generated grayscale image and the high-frequency information as the input of a coloring network; and S4, training the decoloring network and the coloring network in an end-to-end form. The invention relates to the field of image processing, in particular to a frequency regularization reversible grayscale technology based on priori knowledge. The technical problem to be solved by the invention is to provide a frequency regularization reversible gray technology based on priori knowledge, and aims to improve the visual quality of a reconstructed color image.
Need to check novelty before this filing date? Find Prior Art

Citation Information

Patent Citations

  • Reversible grey-scale map algorithm based on L0 regular term and computing device

    CN111696026A

  • Image automatic coloring method and system based on coloring overflow constraint

    CN112991371A