研究目的
To develop a dual pixel-wavelet domain deep CNNs-based soft decoding network for JPEG-compressed images to improve the quality of compressed images without changing codec or introducing extra coding bits.
研究成果
The DPW-SDNet effectively reduces compression artifacts in JPEG-compressed images by leveraging dual pixel-wavelet domain deep CNNs. It outperforms several state-of-the-art algorithms in both objective assessment scores and visual quality. Future work includes extending the framework to other image compression standards and restoration problems.
研究不足
The DPW-SDNet requires training dedicated models for different compression quality factors (QFs), which may limit its flexibility in practical applications where the QF is unknown.
1：Experimental Design and Method Selection
The methodology involves designing a dual-branch deep CNN that performs restoration in both the pixel domain and wavelet domain. The pixel domain branch uses downsampled versions of the compressed image, while the wavelet domain branch uses DWT coefficients.
2：Sample Selection and Data Sources
The publicly available imageset BSD-S500 is used for training, with data augmentation (rotation and downsampling) to generate more training images.
3：List of Experimental Equipment and Materials
GeForce GTX 1080 Ti GPU for training, MATLAB JPEG encoder for generating compressed images.
4：Experimental Procedures and Operational Workflow
The compressed image is downsampled for the pixel domain branch and transformed using DWT for the wavelet domain branch. Both branches process their inputs through deep CNNs, and their outputs are combined to produce the final result.
5：Data Analysis Methods
Performance is evaluated using PSNR, SSIM, and PSNR-B metrics on benchmark datasets Classic5 and LIVE1.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
