摘 要: 为解决真实场景图像采集过程中受随机噪声影响导致采集质量下降的问题,提出一种基于改进KSVD(K-Singular Value Decomposition)的稀疏表示算法。通过无逆稀疏贝叶斯学习(Inverse Free Sparse Bayesian Learning,IFSBL)优化稀疏编码,提高字典原子利用率,让更多的字典原子信号参与到字典更新的过程中,在增强字典表达能力的同时,使稀疏表示的准确性更高;使用IFSBL-KSVD算法对含噪声的图像进行稀疏表示去噪,实验采用通用的图像数据集Set12进行测试,结果表明所提算法的PSNR相比于KSVD的PSNR提升了0.5 dB,能有效提升实际场景中采集图像的质量。 |
关键词: KSVD;IFSBL;稀疏表示;字典学习;图像去噪 |
中图分类号: TP391.1
文献标识码: A
|
基金项目: 国家自然科学基金项目(U1709219,61601410);浙江省科技厅重点研发计划项目(2022C01079);2021年产业技术基础公共服务平台项目(2021-0174-1-1) |
|
Sparse Representation Denoising Algorithm Based on Improved KSVD |
CHEN Pinwei1, LV Wentao1, HAN Tongpeng2, YE Dan3
|
(1.School of In f ormation Science and Engineering, Zhejiang Sci-Tech University, Hangzhou 310018, China; 2.Zhejiang Huanzhi Yunchuang Technology Co., Ltd., Hangzhou 311121, China; 3.Hangzhou Shuye Technology Co., Ltd., Hangzhou 310018, China)
chenpinwei304@163.com; alvinlwt@zstu.edu.cn; hantongpeng83@163.com; lwt93@163.com
|
Abstract: To address the issue of decreased image quality in real-world image acquisition processes due to random noise interference, a sparse representation algorithm based on improved KSVD ( K-Singular Value Decomposition) is proposed. By optimizing sparse coding through Inverse-free Sparse Bayesian Learning ( IFSBL), the utilization efficiency of dictionary atoms is enhanced, allowing more dictionary atom signals to participate in the dictionary update process. This enhances the dictionary 's expressive power and increases the accuracy of sparse representation. The IFSBL-KSVD algorithm is used to denoise images containing noise, and the common image dataset Set12 is used for experiment. Experiment results show that the proposed algorithm improves PSNR (Peak Signal-to-Noise Ratio) by 0.5 dB compared to KSVD, effectively enhancing the quality of image acquisition in practical scenarios. |
Keywords: KSVD; IFSBL; sparse representation; dictionary learning; image denoising |