Table 1

Lists the commonly used spatially varying PSF deconvolution algorithms, all of which belong to the field of computational imaging techniques."

Table 2

Time Consumption (in second) per Iteration[15] for Five Image Stacking Algorithms Applied to Three Mock Samples with Different Sizes, Exposure Counts, and Upsampling Rates β. It is evident that the latter three algorithms, which deliver better results, are more computationally intensive."

Fig.1

Flowchart of the UPDC Forward Modeling Algorithm. The yellow section represents telescope imaging, while the blue section represents simulation imaging and comparative calculations. The high-sampling-rate PSF convolution and resampling modules in the flowchart are the most time-consuming. Forward modeling aligns best with the imaging logic of telescopes, delivering excellent results but requiring substantial computational resources."

Fig.2

Technical Roadmap of Multi-Exposure Spatially Varying PSF Deconvolution Stacking and Domestic Heterogeneous Acceleration."

Fig.3

The 4 Principal Components Used in the Simulation. All components are represented by Sersic functions, which can characterize galaxies, but each has a different orientation."

Fig.4

The 4 Coefficient Fields Corresponding to the 4 Principal Components Used in the Simulation. The coefficient fields exhibit significant variations."

Fig.5

Convolution Effect of Spatially Varying PSF. The panel 5a shows the final spatially varying PSF synthesized from 4 principal components modulated by their corresponding coefficient fields. The panel 5b is one of the 4 exposures taken by JWST-NIRCam in the F444W band, targeting the strong-lensing galaxy M0138. The panel 5c is the convolution result of the left image and the middle image."

Fig.6

The stacking results of spatially varying PSF deconvolution for multiple exposures. Panel 6a shows the direct Drizzle stacking of four multi-exposure images (as in Figure 5b), serving as a comparative reference. Panel 6b presents the Drizzle stacking of the blurred versions of these four multi-exposure images. Panel 6c displays the output from the 20th iteration of our new algorithm, "spatially varying PSF deconvolution stacking for multiple exposures." Panel 6d shows the output from the 60th iteration of the new algorithm."

Table 3

Comparison of Flux Conservation. The central peak flux represents the maximum flux value at the center of the point source, while the total flux within the region is the sum of fluxes in the defined square area. The flux of the point source is set to 10,000, meaning the ideal total flux within the region is 10,000."

Fig.7

The new algorithm was applied to the RGB image of the parallel field (Module-A) of galaxy cluster SMACSJ0723. The left image represents the output from the JWST official pipeline, while the right image shows the output after 10 iterations of the new algorithm. The right image is sharper, with higher signal-to-noise ratio and resolution. The galaxy marked by tag-A appears narrower after processing by the new algorithm, indicating a reduction in PSF effects. The substructure details of the two galaxies marked by tag-B are more clearly distinguishable after processing. The contamination from the neighboring galaxy's diffraction spikes, marked by tag-C, is significantly suppressed by the new algorithm."

Table 4

Statistics of Computation Time. The numbers in the table are averaged over 50 rounds of computation."