Fig. 1. Polychromatic diffraction demonstration. (a) Diffraction setup. (b) Sample image. (c) FT of (b). (d) Obtained through zero-padding around (b). (e) FT of (d). (f) Obtained through cropping (e).

Fig. 2. A2λ0 operations via FTM and scaling method in Ref. [23]. (a) Original diffraction pattern. (b) FT of (a). (c) Obtained by zero-padding around (b). (d) IFT of (c). (e) Obtained by cropping (d). (f) Diffraction pattern derived from the scaling method. (g), (h) FT of (e), (f). (i), (j) Image reconstructed from (e), (f).

Fig. 3. GM-FTM with simulated ultra-broadband diffraction. Color bars are consistent across all figures. (a) Spectra used in simulations. The wavelengths are in arbitrary units and nanometers for model sources and experimental attosecond laser source, respectively. (b) Object intensity. (c) FT of (b). (d1)–(d4), (e1)–(e4), (f1)–(f4) Broadband pattern, monochromatic pattern determined via GM-FTM, the reconstructed image and difference between reconstructed image and the ground truth for the model continuous, model comb-like, and experimental attosecond pulse spectra, respectively.

Fig. 4. Spectral characteristics of employed lasers and quantum efficiency of the CMOS sensor.

Fig. 5. Ti:sapphire laser post-supercontinuum generation results. αc is approximately three for group 7. (a), (b) Diffraction pattern and reconstructed image for He-Ne laser. (c), (d) Diffraction pattern and direct CDI reconstructed image for Ti:sapphire laser. (e1)–(e3) Monochromatized pattern recovered by GM-FTM, reconstructed image, and difference between (e2) and (b), respectively.

Fig. 6. Resolution comparison of imaging with monochromatic and broadband sources. The image intensity profiles in (a) correspond to the red line in (b).

Fig. 7. Supercontinuum light source results. αc is approximately six. (a), (b) Diffraction pattern and reconstructed image for He-Ne laser. (c), (d) Diffraction pattern and direct CDI reconstructed image for supercontinuum laser. (e1)–(e3) Monochromatized pattern recovered by GM-FTM, reconstructed image, and difference between (e2) and (b), respectively.

Fig. 8. Experimental results of spectrum reconstruction. (a) Sample image reconstructed by CDI algorithms based on the pattern (b), acquired using a He-Ne laser. (c) Broadband diffraction pattern obtained using a continuum source with shorter wavelength components filtered out. The coefficients ai are derived from (b) and (c), as indicated by the red line in (d). The blue line represents values calculated from spectrometer output, adjusted for CMOS response, serving as a comparative metric.

Fig. 9. (a) Algorithm demonstration using broadband diffraction data. (b) Image reconstructed directly from (a). (c) Illumination spectrum and sampling strategies. (d) Converging speed comparison of direct gradient method and momentum method with k=0.8 and k=0.6, respectively. (e1)–(e4) Monochromatic pattern reconstructed using GM-FTM with the smallest step, four points, three points, and two points spectrum sampling strategy. (f1)–(f4) Image reconstructed from (e1)–(e4).

Fig. 10. Noise impact on the GM-FTM reconstruction.

Fig. 11. Microscopic images of the samples.

Fig. 12. Ti:sapphire laser post-supercontinuum generation results using the method proposed by Ref. [23]. (a) Monochromatized pattern. (b) Reconstructed image. (c) Difference between reconstructed image and Fig. 5(b).