To tackle the limitations of existing image compressive sensing reconstruction methods, such as the separation of sparsity constraints and shrinkage mechanisms, fixed sparsity and shrinkage threshold values, absence of rate-distortion optimization, and poor adaptability at low sampling ratios, along with common deep network defects including vanishing gradients, unstable training, weak anti-noise performance and limited generalization ability for practical scenarios, the work aims to propose the S&SCCA, an image compressive sensing reconstruction model based on joint sparsity and shrinkage collaborative optimization. The constructed model achieved breakthroughs in both processing mechanism and modeling paradigm. It unified sparsity constraints and shrinkage mechanisms within a single loss function, breaking the conventional practice where the two modules operated independently. The introduced adaptive sparse coefficients enabled dynamic adjustment of sparsity intensity according to the complexity of image features, while the defined learnable shrinkage thresholds allowed end-to-end optimization of threshold values varying with sampling ratios. Moreover, a multi-objective joint loss control strategy was proposed to integrate the collaborative optimization of reconstruction, sparsity, shrinkage, bit rate and regularization into an integrated optimization framework. Relevant experiments demonstrated that the collaborative optimization mechanism, adaptive strategy, residual shrinkage structure and rate-distortion optimization could progressively boost image reconstruction accuracy, stabilize model training, and satisfy the requirements of high compression ratios under low sampling rates. At the low sampling ratio of 0.1, the model achieved a PSNR of 39.05 dB, an SSIM of 0.84, a training-validation loss gap of 0.06, a VIF of 0.73 and an EPI of 0.82, with nearly 40% faster convergence. It delivered favorable compression efficiency and inference speed, and exhibited outstanding and stable overall performance in practical engineering scenarios. The S&SCCA model effectively addresses core challenges in image compression coding and decoding reconstruction via collaborative sparsity-shrinkage optimization, adaptive dynamic adjustment, lightweight residual architecture, and joint rate-distortion optimization. It possesses favorable generalization capability under complex environments and noise interference, and holds practical engineering application value. This method can serve as an effective solution for image compressive sensing, lightweight reconstruction, and anti-interference transmission.
Key words
image compressive sensing /
joint sparsity and shrinkage collaborative optimization /
learnable threshold /
residual shrinkage /
rate-distortion optimization /
robust reconstruction
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
References
[1] CHEN J, ZHU Y T, HUANG W, et al.Scale-Adaptive Asymmetric Sparse Variational AutoEncoder for Point Cloud Compression[J]. IEEE Transactions on Broadcasting, 2024, 70(3): 884-894.
[2] LIU J H, YUAN F, XUE C, et al.An Efficient and Robust Underwater Image Compression Scheme Based on Autoencoder[J]. IEEE Journal of Oceanic Engineering, 2023, 48(3): 925-945.
[3] 周开军, 廖婷, 谭平, 等. 图像压缩技术研究综述[J]. 计算机科学与探索, 2025, 19(7): 1699-1728.
ZHOU K J, LIAO T, TAN P, et al.Review of Research on Image Compression Techniques[J]. Journal of Frontiers of Computer Science & Technology, 2025, 19(7): 1699-1728.
[4] 岳爽, 陈喆, 殷福亮. 极低比特率图像压缩技术综述[J]. 数据采集与处理, 2025, 40(1): 102-116.
YUE S, CHEN Z, YIN F L.Review of very Low Bitrate Image Compression Techniques[J]. Journal of Data Acquisition and Processing, 2025, 40(1): 102-116.
[5] 徐雯, 于瓅. 基于迭代收缩阈值与深度学习的压缩感知图像重构网络[J]. 计算机工程与科学, 2025, 47(3): 485-493.
XU W, YU L.A Compressive Sensing Image Reconstruction Network Based on Iterative Shrinkage Thresholding and Deep Learning[J]. Computer Engineering & Science, 2025, 47(3): 485-493.
[6] 张俊洋, 明镝. 基于自编码器和稀疏学习的深度鲁棒图像重构[J]. 计算机技术与发展, 2025, 35(10): 43-52.
ZHANG J Y, MING D.Deep Robust Image Reconstruction Based on Autoencoder and Sparse Learning[J]. Computer Technology and Development, 2025, 35(10): 43-52.
[7] 郝宇星, 刘庆强. 基于收缩自编码器和局部保持投影的机械故障特征提取[J]. 现代制造工程, 2025(2): 84-93.
HAO Y X, LIU Q Q.Mechanical Fault Feature Extraction Based on Contractive Auto-Encoder and Locality Preserving Projections[J]. Modern Manufacturing Engineering, 2025(2): 84-93.
[8] 吴亚丽, 冯梦琦, 王君虎, 等. 变分模态分解和自适应稀疏自编码器的故障诊断模型[J]. 机械科学与技术, 2025, 44(9): 1603-1611.
WU Y L, FENG M Q, WANG J H, et al.Fault Diagnosis Method Combining Variational Mode Decomposition with Adaptive Sparse Autoencoder[J]. Mechanical Science and Technology for Aerospace Engineering, 2025, 44(9): 1603-1611.
[9] 佘丁辰, 王威, 王加琪, 等. 基于收缩自编码器的无人机GPS欺骗攻击协同检测方法[J]. 信号处理, 2024, 40(4): 706-718.
SHE D C, WANG W, WANG J Q, et al.Collaborative Detection Method of UAV GPS Spoofing Attack Based on Shrink Autoencoder[J]. Journal of Signal Processing, 2024, 40(4): 706-718.
[10] 潘雪娇, 董绍江, 周存芳, 等. 基于深度卷积自编码器和多尺度残差收缩网络的滚动轴承寿命状态识别[J]. 重庆交通大学学报(自然科学版), 2024, 43(5): 124-132.
PAN X J, DONG S J, ZHOU C F, et al.Rolling Bearing Life State Recognition Based on Deep ConvolutionalAutoencoder and Multi-Scale Residual Shrinkage Network[J]. Journal of Chongqing Jiaotong University (Natural Science), 2024, 43(5): 124-132.
[11] 方芯, 王伟, 贺丽君, 等. 结构-语义-残差协同的图像压缩框架:面向隐私保护与高保真重建[J]. 西安交通大学学报, 2026, 60(7): 185-195.
FANG X, WANG W, HE L J, et al.A Structure-Semantics-Residual Collaborative Image Compression Framework: Towards Privacy Protection and High-Fidelity Reconstruction[J]. Journal of Xi'an Jiaotong University, 2026, 60(7): 185-195.
[12] 张璐亨, 王文煜, 丁丹丹.面向端到端图像压缩的可分级码率控制方法[J/OL]. 北京航空航天大学学报, 2026, 52(0): 1-12.[2026-03-12]. https://doi.org/10.13700/j.bh.1001-5965.2024.0461.
ZHANG Luheng, WANG Wenyu, DING Dandan.A Hierarchical Rate Control Method for End-to-End Image Compression[J/OL]. Journal of Beijing University of Aeronautics and Astronautics, 2026, 52(0): 1-12. [2026-03-12]. https://doi.org/10.13700/j.bh.1001-5965.2024.0461.
[13] 付邦劭, 沈圣, 杨本植, 等. 控光器件-图像编码联合优化的集成成像角分辨率提升方法[J]. 光学学报, 2026, 46(11): 1111008.
FU B S, SHEN S, YANG B Z, et al.Angular Resolution Enhancement Method of Integral Imaging System Based on Joint Optimization of Optical Control Device and Image Coding[J]. Acta Optica Sinica, 2026, 46(11): 1111008.
[14] CAROLLO M, FERRERO-JAURRIETA M, PAIVA R, et al.Enhancing Feature Compression and Reconstruction in Time-Series and Image Domains with Pseudo-Overlap and Pseudo-Grouping Pooling Functions[J]. Knowledge-Based Systems, 2026, 336: 115333.
[15] SHI K J, SHI C P, SUN W W, et al.A Spectral Difference Preservation Network Based on Mamba Pyramid for Hyperspectral Image Compression[J]. Pattern Recognition, 2026, 175: 113088.
[16] STERN A.Compressing and Expanding Optical Matrix-Vector Multipliers for Enabling Optical Image Encoder-Decoders and Generators[J]. Light: Science & Applications, 2026, 15: 45.
[17] WANG J Y, ZHANG X Y, MA W Z, et al.Understanding Internal Representations of Recommendation Models with Sparse Autoencoders[J]. ACM Transactions on Information Systems, 2026, 44(3): 1-29.
[18] SONG D W, YAN Q R, YANG J, et al.Content-Aware Transformer Network for Image Compressed Sensing[J]. IEEE Internet of Things Journal, 2026, 13(11): 25274-25285.
[19] XIA Y C, ZHOU Y M, WANG J P, et al.DiRIC: Diffusion Prior Refinement for Efficient Low-Rate Image Compression[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2026. DOI: 10.1109/TPAMI.2026.3704972.
[20] GONZÁLEZ-DE-REGÀS J, MIJARES S, BARTRINA- RAPESTA J, et al. Robustness of Lossy Multispectral Compression to Simulated Instrumental Noise: A Comparative Study[J]. IEEE Geoscience and Remote Sensing Letters, 2025, 22: 5003305.
[21] 王瑞青, 何昆仑, 陈华, 等. 面向医学影像细微特征的低损耗压缩编码算法的研究与应用[J]. 解放军医学院学报, 2025, 46(10): 982-987.
WANG R Q, HE K L, CHEN H, et al.Research and Application of Low-Loss Compression Coding Algorithms for Fine Features in Medical Imaging[J]. Academic Journal of Chinese Pla Medical School, 2025, 46(10): 982-987.
PDF(7840 KB)
