Image reconstruction in graphic design based on Global residual Network optimized compressed sensing model

Area of Science:

• Computer Vision
• Image Processing
• Signal Processing

Background:

• Traditional compressed sensing (CS) models face challenges with information degradation and distortion in graphic design applications.
• Optimizing CS models is crucial for accurate image reconstruction from limited data.

Purpose of the Study:

• To optimize the traditional compressed sensing (CS) model for graphic design to mitigate information degradation and distortion.
• To enhance the accuracy of image reconstruction, particularly for high-frequency details, using an improved CS approach.

Main Methods:

• A co-reconstruction group strategy was developed using compressed observations of local image blocks.
• An initial reconstruction of similar image blocks was performed, followed by channel stitching.
• A global residual network with a non-local feature adaptive interaction module was employed for enhanced local feature reconstruction.

Main Results:

• The optimized CS model achieved solution space constraints for reconstructed images at low sampling rates.
• High-frequency information within images was effectively reconstructed, significantly improving overall image reconstruction accuracy.
• The proposed method demonstrates superior performance in preserving image details compared to traditional CS models.

Conclusions:

• The developed co-reconstruction group strategy and global residual network effectively address information degradation in CS.
• This optimized CS model offers a viable solution for high-fidelity image reconstruction in graphic design, even with minimal sampling.
• The findings contribute to advancing image reconstruction techniques in digital imaging and graphic design fields.