Fig. 1. The schematic framework of spatial-frequency Trust-Region algorithm by FMM.

Fig. 2. The structural schematics of the optimized and fabricated metasurfaces, and the transfer function curves (both simulated and experimental) of high-contrast and high-resolution differentiation metasurfaces. (a), (b) Structural schematics of optimized high-contrast (a) and high-resolution (b) differentiation metasurface. (c), (d) SEM images of fabricated high-contrast (c) and high-resolution (d) differentiation metasurface. (e), (f) Simulated and experimental SFTF curves of both kinds of metasurfaces. (g) Experimental set-up of SFTF curve measurement.

Fig. 3. Experimental set of resolution measurements and experimental results. (a) The experimental set-up of edge detection implemented by the fabricated differentiation metasurfaces. (b)–(d) Input images of resolution test charts in different sizes. (e)–(j) Output images of high-contrast and high-resolution differentiation metasurfaces. (k)–(m) Intensity profiles at the position of white dashed line in (b), (e), and (h).

Fig. 4. Edge detection of cells. (a)–(d) Input images of cells: onion epidermis (a), tomato cells (b), mouse hepatocytes (c), and mouse spleen cells (d). (e)–(l) Output images of high-contrast (e)–(h) and high-resolution (i)–(l) differentiation metasurface for (a)–(d), respectively.

Fig. 5. Edge detection of dynamic paramecia images and two-step edge detection framework of dynamic biomedical images. (a) Paramecia image observed by microscope. (b), (c) Edge detection results of paramecia images by high-contrast (b) and high-resolution (c) differentiation metasurface (see Visualization 1). (d) Schematic of the two-step edge detection framework of dynamic biomedical images. (e), (f) Frames of the observation video of Step 1 (e) and Step 2 (f) for an alive Brachionus (see Visualization 2).