Imaging Through Plasmonic Nanoparticles

Imaging through plasmonic nanoparticles

Mehbuba Tanzida, Ali Sobhania, Christopher J. DeSantisb, Yao Cuib, Nathaniel J. Hoganc, Adam Samaniegoa, Ashok Veeraraghavana, and Naomi J. Halasa,b,c,1

aDepartment of Electrical and Computer Engineering, Rice University, Houston, TX 77005; bDepartment of Chemistry, Rice University, Houston, TX 77005; and cDepartment of Physics and Astronomy, Rice University, Houston, TX 77005

Contributed by Naomi J. Halas, March 31, 2016 (sent for review March 2, 2016; reviewed by Alan C. Bovik and Brendan G. DeLacy) The optical properties of metallic nanoparticles with plasmon image are attenuated by transmission through the nanoparticle- resonances have been studied extensively, typically by measuring based medium. These two approaches provide insight toward the transmission of light, as a function of wavelength, through a understanding how our ability to perceive and resolve an image nanoparticle suspension. One question that has not yet been transmitted through a nanoparticle-based medium is dependent addressed, however, is how an image is transmitted through such upon the specific properties of the nanoparticles themselves in a suspension of absorber-scatterers, in other words, how the addition to the spatial frequencies of the image. various spatial frequencies are attenuated as they pass through The problem under investigation is illustrated schematically in the nanoparticle host medium. Here, we examine how the optical Fig. 1. A medium consisting of suspended plasmonic nanoparticles properties of a suspension of plasmonic nanoparticles affect the is placed between an object and the image plane of an observer, transmitted image. We use two distinct ways to assess transmitted or, alternatively, an imaging system. The extinction spectrum of image quality: the structural similarity index (SSIM), a perceptual the plasmonic medium alone provides valuable but incomplete distortion metric based on the human visual system, and the information concerning the quality of the transmitted image. The modulation transfer function (MTF), which assesses the resolvable spatial frequencies. We show that perceived image quality, as well spectrum does not provide the specific wavelength-dependent as spatial resolution, are both dependent on the scattering and scattering and absorption properties of the nanoparticles, nor does absorption cross-sections of the constituent nanoparticles. Sur- it account for the fact that different spatial frequencies in an image prisingly, we observe a nonlinear dependence of image quality on are differentially transmitted or attenuated while passing through optical density by varying optical path length and nanoparticle the medium. concentration. This work is a first step toward understanding the Qualitatively, in wavelength regions where the extinction is low, requirements for visualizing and resolving objects through media we would expect the transmitted image to be more easily visible consisting of subwavelength absorber-scatterer structures, an than for wavelength regions of higher extinction. This extinction approach that should also prove useful in the assessment of will be vastly different for different spatial frequencies—higher metamaterial or metasurface-based optical imaging systems. spatial frequencies will invariably be attenuated far more than lower spatial frequencies. To quantitatively assess the correlation light scattering | imaging | nanoparticles | plasmonics | metamaterials between image visibility and nanoparticle optical properties, we acquired transmitted images at discrete wavelengths across the ur understanding of light scattering in complex media is visible spectrum. We then analyzed these images using compu- Obuilt almost exclusively on the interaction of light with tational image-processing tools to correlate nanoparticle optical broadband low-loss scatterers, characteristic of many natural properties with human-eye image perception quantitatively. systems such as animal tissue, fog, or sea spray (1–6). In contrast, metallic nanoparticles possess plasmon resonances with strongly Significance frequency-dependent absorption and scattering cross-sections, where the resonant frequency is controlled by nanoparticle size, How is an image transmitted through a material consisting of – shape, and local dielectric environment (7 9). This characteristic subwavelength structures? We use two distinct methods to enables the predictive design and fabrication of nanoparticles obtain quantitative answers to that question. The first, struc- and nanoparticle-based media with specific absorption and tural similarity index, is a method related to human perception, scattering properties at precise wavelengths of choice throughout initially developed to quantify transmitted image quality fol- the UV, visible, and infrared regions of the spectrum (10–16). lowing image compression in digital image transmission sys- Since their use in antiquity as the vivid colorants in stained glass tems. The second method treats the medium as an optical windows, this frequency dependence has been of primary interest. component itself, where we determine the spatial frequency In the context of modern optics, plasmon-resonant lineshapes are content of the image transmitted by the medium. This study studied extensively through spectroscopic measurements. How- opens the door to analyzing images transmitted through par- ever, the problem of resolving an image through a suspension of ticulate media that absorb and/or scatter light, which applies plasmonic nanoparticles, for example, dispersed in an otherwise generally to imaging systems whose components are composed transparent solid or liquid medium, or patterned onto a trans- of subwavelength structures, such as those composed of random parent substrate, has yet to be investigated. Here, we examine how particulates or nanoengineered flat-optics metasurface lenses. the properties of plasmonic nanoparticles in dilute suspension modify a transmitted image. To do so, we use two very different Author contributions: M.T., A. Sobhani, A.V., and N. J. Halas designed research; M.T. and A. Sobhani performed the experiments; C.J.D. synthesized nanorods for the experiments; strategies. First, we determine the Structural SIMilarity index M.T., Y.C., N. J. Hogan, and A. Samaniego analyzed data; and M.T., A.V., and N. J. Halas (SSIM) of the plasmonic medium-transmitted image relative to wrote the paper. the original image (17). This technique allows us to evaluate how Reviewers: A.C.B., University of Texas; and B.G.D., US Army Edgewood Chemical an image is distorted upon transmission by applying a quantita- Biological Center. tive metric related directly to human visual perception. Second, The authors declare no conflict of interest. we consider the plasmonic nanoparticle medium as if it were Freely available online through the PNAS open access option. an optical component and evaluate the medium’s modulation 1To whom correspondence should be addressed. Email: [email protected]. transfer function (MTF) (18). This approach provides quantita- This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10. tive information concerning how various spatial frequencies in an 1073/pnas.1603536113/-/DCSupplemental.

5558–5563 | PNAS | May 17, 2016 | vol. 113 | no. 20 www.pnas.org/cgi/doi/10.1073/pnas.1603536113 Downloaded by guest on October 1, 2021 wavelength-selective bandpass filters (FWHM, ±10 nm; FB series; Thorlabs) spanning the wavelength range of 450–900 nm were used. For imaging, a CCD camera (PIXIS 400; Princeton Instruments) sensitive in the 400 to 1,100 nm wavelength range was used. The object was a 1951 US Air Force (USAF) resolution target conforming to the military standard (MIL-STD- 150A) (SI Appendix,Fig.2). Nanoparticle suspension in a quartz cuvette was placed directly in front of the camera so that the suspension influenced only the target image and not the illumination source. A suspension of plasmonic nanoparticles was designed and prepared with spectrally distinct transmission maxima and minima (Fig. 2). The suspension consisted of a mixture of Au nanorods of two different sizes and aspect ratios corresponding Fig. 1. Imaging through a medium of plasmonic nanoparticles. An image of A the object (Left), under white light illumination, is transmitted through a to two spectrally distinct dipolar plasmon modes (Fig. 2 ). One suspension of nanoparticles with a wavelength-dependent extinction spec- type of Au nanorods were 20 nm × 45 nm in size with a dipolar trum. In this case, the suspension has a transmission maximum at green wavelength (G) and transmission minima at blue (B) and red (R) wavelengths. Here, we examine how the observer perceives the transmitted image in these different wavelength regions and how the absorption and scattering properties of the nanoparticles affect the perceived image.

The experiment consisted of a bright field Kӧhler illumination setup for imaging an object through plasmonic nanoparticle suspensions (details shown in SI Appendix,Fig.1). A fiber optic white light illumination source (MI-150; Edmund Optics) with a series of ENGINEERING

Fig. 2. Mixture of two nanorod solutions for imaging through a complex resonant medium. (A) Measured extinction spectrum of the mixture of Fig. 3. SSIM quality metric for images obtained through nanoparticle 20 nm × 45 nm and 30 nm × 120 nm Au nanorod solutions. (B) Measured solution. (A) Normalized extinction spectrum of nanorod-mixture sus- extinction spectrum of 20 nm × 45 nm nanorod solution. (B, Inset) TEM image. pension, indicating wavelengths of transmitted images. (B) Normalized (C)Measuredextinctionspectrumof30nm× 120 nm Au nanorod solution. images of one set of vertical bars in the USAF resolution target obtained (C, Inset) TEM image. (D) Calculated absorption (blue), scattering (red), and through the nanorod suspension from 450 to 900 nm wavelengths at 50

extinction (black) cross-sections (in H2O)ofa45nm× 20 nm Au nanorod. nm wavelength intervals. (C) USAF target images obtained through H2O. (E)A30nm× 120 nm Au nanorod. (D)CalculatedSSIMandMS-SSIMfortheseriesofimagesshowninB.

Tanzid et al. PNAS | May 17, 2016 | vol. 113 | no. 20 | 5559 Downloaded by guest on October 1, 2021 Fig. 4. MTFs of the imaging system with the mixed-nanorod suspension included. (A, Top) Extinction spectrum of mixed-nanorod suspension, where the 650 nm (blue), 750 nm (green), and 850 nm (red) illumination wavelengths are shown. (A, Middle and Bottom) Normalized images of 17.96 to 228.1 lp/mm in the USAF

resolution target through mixed-nanorod suspension, with 2.5-mm optical path length (Middle) and through H2O as a control (Bottom). (B–D) MTF of the imaging system with H2O and mixture of 20 nm × 45 nm and 30 nm × 120 nm at 650 nm (B), 750 nm (C), and 850 nm (D) wavelengths. The cutoff MTF of 0.05 is shown by the black dashed line in each figure. Dotted portions of the MTF curves, required to determine the cutoff spatial frequency, are linearly interpolated data. (E)Cutoff

spatial frequencies through H2O and the mixed-nanorod suspension, for the three frequencies of interest.

resonance wavelength of 637 nm (Fig. 2B). The second type of practice, an SSIM of 0 indicates no similarity between the trans- Au nanorods were 30 nm × 120 nm in size with a dipolar reso- mitted and the original image, and an SSIM of 1 indicates no nance wavelength of 877 nm (Fig. 2C). Finite-difference time detectable difference between the original and the transmitted domain (FDTD) calculations of Au nanorods with these di- image. Additionally, multiscale (MS)-SSIM (20) can be calculated mensions indicate that absorption dominates the cross-section of to take into account image details, such as the sampling density the smaller nanorods (Fig. 2D), whereas for the larger nanorods, of the image signal, the distance from the image plane to the scattering and absorption contribute equally (Fig. 2E). Mixing observer, and the perceptual capability of the observer’svisual the nanorod solutions results in an extinction minimum at 770 nm system at different resolutions (more details are provided in the and two maxima at 637 and 877 nm, which frame the transmis- SI Appendix). sion window. Normalized monochromatic images of a set of For the transmitted images displayed in Fig. 3B and using the vertical bars [12.7 line pairs/mm (lp/mm)] on the USAF resolu- images shown in Fig. 3C as the original images, we calculated the tion target were obtained after transmission through the mixed- SSIM (red) and MS-SSIM (blue) values for each transmitted nanorod suspension (Fig. 3). The extinction spectrum of the image, as shown in Fig. 3D. We can see that these values cor- suspension and the specific illumination wavelengths used for relate well with the quality of the images as perceived by the imaging, which ranged from 450 to 900 nm at 50 nm wavelength human eye. The SSIM and MS-SSIM follow the transmission intervals, are shown (Fig. 3A). The image is transmitted clearly spectrum of the solution to some extent. MS-SSIM shows a within the transmission window near 750 nm and becomes in- slightly more uniform spectral distribution compared with SSIM. creasingly distorted for illumination wavelengths detuned from However, both deviate from the spectrum at certain wavelengths. the 750 nm transmission window (Fig. 3B). Imaging through the For example, even though the extinction maxima near 650 and aqueous solution (Fig. 3C) shows that the observed image dis- 850 nm have comparable optical densities, the image obtained at tortion is attributable to transmission through the mixed nano- 850 nm wavelength (with SSIM 0.094 and MS-SSIM 0) has sub- particle suspension. To quantitatively analyze and assess the quality stantially greater image distortionthantheimageobtainedat650nm of the transmitted image, we apply a quality metric known as the wavelength (with SSIM 0.285 and MS-SSIM 0.131). The differ- SSIM. Unlike traditional methods for evaluating image quality, such ence between the change in image distortion at 650 and 850 nm as mean squared error (MSE) or peak signal-to-noise ratio (PSNR) can be attributed to the difference in the ratio of scattering to (SI Appendix,Fig.3), SSIM is not biased toward oversmoothed or absorption cross-sections for the two nanorod sizes in the mixed- blurry results but has been proven to be consistent with human nanorod suspension: for the 20 nm × 45 nm nanorods, this ratio visual perception (19). The human visual system is highly adaptive is 0.11, and for the 30 nm × 120 nm nanorods, this ratio is 1.04. for extracting structural information from a scene: this is taken Because the smaller nanorods with their dipolar plasmon into account in the definition of the SSIM metric [computational near 650 nm scatter less light than those with their resonance details are given in the SI Appendix, and we refer the reader to the near 850 nm and because absorbed light merely reduces the original publication (17) for details regarding the SSIM metric]. In brightness of the image and not the contrast, there is less

5560 | www.pnas.org/cgi/doi/10.1073/pnas.1603536113 Tanzid et al. Downloaded by guest on October 1, 2021 Fig. 5. Variation of image quality with optical path length through nanoparticle suspensions. (A) Normalized extinction spectra of nanorod-mixture window (2.5-mm optical path length) (Upper) along with normalized images of 17.96–228.1 lp/mm in the USAF resolution target through nanorod-mixture sus-

pensions with optical path lengths of 2.5, 5, and 10 mm (Lower, top three rows) and through H2O(Lower, bottom row) shown from 450 to 900 nm wavelengths with a 50 nm wavelength interval. (B and C) Calculated SSIM (B) and cutoff spatial frequency (C) for the same images obtained through nanorod-mixture suspension for three different optical path lengths (2.5, 5, and 10 mm).

perceived image distortion at the 650 nm illumination wave- 4 B–D. At 750 nm, the MTF through the mixed-nanorod sus- length. Conversely, the larger nanorods with their dipolar res- pension is almost equal to the MTF through H2O. However, at onance near 850 nm have a larger relative scattering cross- the resonance peaks (650 and 850 nm), the MTF through the section, which distorts the image, because scattering results in mixed-nanorod suspensions is substantially reduced (Fig. 4 B redirected photons that, although detectable, have lost in- and D). Although the optical density of the mixture is similar formation regarding the object, resulting in a distorted image. at both resonance peaks, the absorption and scattering cross- This result indicates that the absorption and scattering prop- sections are substantially different. The MTF determined at erties of the constituent nanoparticles of a plasmonic medium 850 nm, where scattering is more dominant, is reduced more directly affect the quality of a transmitted image. than the MTF at 650 nm. The MTF cutoff for the average Another method to assess the quality of an image transmitted human visual system, 0.05, is shown by the black dashed lines through a plasmonic medium is to consider the medium to be in Fig. 4 B–D.InFig.4E, we see that at the transmission an additional component of the imaging system and determine maximum, the cutoff spatial frequency is the same for the the MTF of that component (Fig. 4). The MTF describes the mixed-nanorod suspension as it is in H2O, meaning that the maximum resolution of an imaging system as a function of maximum resolvable spatial frequency through the nano- transmittable spatial frequencies (lp/mm), indicating the mini- particle suspension is unaltered at this wavelength. The cutoff mum-sized feature of an object resolvable in a transmitted resolution is much lower at the longer wavelength resonance, image (18, 21) (details are provided in the SI Appendix). MTF is where scattering is stronger (850 nm), than for the shorter

also a normalized metric, whose amplitudes range from 0 to 1, wavelength, predominantly absorptive resonance (650 nm). ENGINEERING where 0.05 is the threshold for resolvability for average human This finding clearly indicates that scattering, rather than ab- observers (22). To determine the MTF of the mixed-nanorod sorption, is the dominant light interaction mechanism that suspension, we imaged a specific portion (17.96 to 228.1 lp/mm) limits the size of the smallest resolvable feature when imaging of the USAF resolution target through the suspension. Nor- through plasmonic media. malized images through the mixed-nanorod suspension, along The total optical density of a plasmonic nanoparticle sus- with images transmitted through an aqueous medium as a pension also affects the quality of the transmitted image. One control, at the wavelength of maximum transmission (750 nm) can vary the optical density of a plasmonic nanoparticle sus- and at the two resonance peaks (650 and 850 nm) are shown in pension by varying either the optical path length or the con- Fig. 4A. The MTF from each transmitted image is shown in Fig. centration of nanoparticles in suspension. First, we varied the

Tanzid et al. PNAS | May 17, 2016 | vol. 113 | no. 20 | 5561 Downloaded by guest on October 1, 2021 to 2.6 × 1010 particles/cm3 and for the 20 nm × 45 nm Au nanorod solutions, the concentration was varied from 7.7 × 1010 to 4.6 × 1011 particles/cm3. We performed imaging at 600 nm wavelength, avoiding the interband transition of gold at 520 nm wavelength (23). We calculated the SSIMs for the images obtained through these nanorod and nanosphere solutions at different concentrations, which are plotted against the corresponding optical densities in Fig. 6. The imaging threshold observable in Fig. 5 can also be observed here as a nonlinear dependence of the SSIM on the optical density of the solution. Below a threshold optical density (<1.5), the image transmits clearly through the nanoparticle solutions with minimum distortion, regardless of the absorption or scattering properties of the plasmonic nanoparticles. However, above a threshold optical density (shown by the gray area in Fig. 6), there is a rapid decrease in image quality, according to the SSIM metric. Additionally, the threshold optical densities, or concentrations, are different for the two nanoparticle solutions. The SSIM drops below 0.2 at a concentration of 1.3 × 1010 particles/cm3 (optical Fig. 6. Variation of image quality with optical density of nanoparticle density 2.3) for the nanosphere solution with larger scattering cross- suspensions as a function of nanoparticle concentration. Measured SSIM for sections and at a concentration of 2.7 × 1011 particles/cm3 (optical USAF target images obtained through suspensions of 100 nm diameter Au density 2.8) for the nanorod solution, which is more absorbing. nanospheres and 20 nm × 45 nm Au nanorods at a 600 nm illumination Although the value of the threshold optical density is also de- wavelength as a function of optical density. pendent upon the specifics of the imaging system, this threshold behavior should be generally observable in any imaging system. optical path length of the nanorod-mixture suspension through Our quantitative study on the influence of plasmonic nano- which objects were imaged (Fig. 5). The normalized images of particles on the visual quality of images transmitted passing 17.96–228.1 lp/mm in the USAF resolution target that were through such nanoparticle-laden media may open a new area of obtained for optical path lengths of 10, 5, and 2.5 mm (Fig. 5A, research to use engineered plasmonic nanoparticles not only for reshaping the light spectrum but also to selectively preserve or Lower, top three rows) and through H2O(Fig.5A, Lower, bottom row) are shown in Fig. 5. The SSIM and cutoff spatial distort a transmitted image for visual or machine detection. frequencies across the spectrum as a function of optical path Among various image quality metrics, SSIM provides realistic length are shown in Fig. 5 B and C. As optical path length is information regarding the clarity or distortion of an image when increased from 2.5 to 5 mm, high image quality as well as spatial transmitted through plasmonic nanoparticles that is directly re- resolution persist within a broad spectral window centered at lated to human perception. On the other hand, MTF evaluates the wavelength of maximal transmission. As we increase opti- the performance of plasmonic nanoparticles by considering them cal path length further (from 5 to 10 mm), image quality and as a part of the imaging system and provides the maximum maximum spatial resolution drop off dramatically. This re- spatial resolution visible through the particular plasmonic sult likely indicates a threshold optical density where optical nanoparticle-based medium. We showed that for equal extinc- densities larger than this value will distort all images be- tion values, plasmonic nanoparticles with higher scattering cross- yond recognition and below this value, there will be minimal sections more effectively distort transmitted images compared image distortion. with nanoparticles with more predominant absorption cross- To further examine this imaging threshold, we determined the sections. Furthermore, we showed that image distortion and SSIM obtained through two different suspensions of individual optical density of the nanoparticle solution have a strongly nanoparticles, this time changing optical density by varying the nonlinear relation, where below a threshold optical density, there concentration of nanoparticles in each suspension. We chose is virtually no image distortion, whereas above this threshold, the 100 nm diameter Au nanospheres and 20 nm × 45 nm Au nanorods image distortion drops dramatically with increasing optical for this study, because they both have similar extinction maxima density. We believe that this initial study will facilitate the as- within the same 600 to 650 nm wavelength window but possess sessment of image visualization through other types of media different absorption and scattering properties (SI Appendix,Fig.4). composed of subwavelength nanostructures with light-scattering Specifically, the ratio of scattering and absorption cross-sections for properties, such as metasurface lenses (24, 25), which can be the nanosphere solution is 40 times larger than the nanorod solu- combined into flat optics-based imaging systems. tion at the 600 nm imaging wavelength. We varied the optical densities of these two plasmonic nanoparticle solutions by varying ACKNOWLEDGMENTS. This work was funded by Robert A. Welch Founda- SI Appendix tion Grant C-1220 and Army Research Office Grant R17830. A.V. was partially their concentrations ( ,Fig.5). The concentration of supported by an Office of Naval Research (ONR) grant on imaging through 9 100 nm diameter Au nanosphere solutions was varied from 1.9 × 10 scattering media.

1. Leith EN, Upatnieks J (1966) Holographic imagery through diffusing media. J Opt Soc 8. Link S, El-Sayed MA (1999) Spectral properties and relaxation dynamics of surface Am 56(4):523. plasmon electronic oscillations in gold and silver nanodots and nanorods. J Phys Chem 2. Freund I (1990) Looking through walls and around corners. Phys A Stat Mech Appl B 103(40):8410–8426. 168(1):49–65. 9. Klar T, et al. (1998) Surface-plasmon resonances in single metallic nanoparticles. Phys 3. Bissonnette LR (1992) Imaging through fog and rain. Opt Eng 31(5):1045–1052. Rev Lett 80(19):4249–4252. 4. Newman JA, Webb KJ (2014) Imaging optical fields through heavily scattering media. 10. Tao A, Sinsermsuksakul P, Yang P (2007) Tunable plasmonic lattices of silver nano- Phys Rev Lett 113(26):263903. crystals. Nat Nanotechnol 2(7):435–440. 5. Studer V, et al. (2012) Compressive fluorescence microscopy for biological and hy- 11. Lee KS, El-Sayed MA (2005) Dependence of the enhanced optical scattering perspectral imaging. Proc Natl Acad Sci USA 109(26):E1679–E1687. efficiency relative to that of absorption for gold metal nanorods on aspect ratio, size, 6. Mosk AP, Lagendijk A, Lerosey G, Fink M (2012) Controlling waves in space and time end-cap shape, and medium refractive index. JPhysChemB109(43):20331–20338. for imaging and focusing in complex media. Nat Photonics 6(5):283–292. 12. Peng S, McMahon JM, Schatz GC, Gray SK, Sun Y (2010) Reversing the size- 7. Maier SA, Atwater HA (2005) Plasmonics: Localization and guiding of electromagnetic dependence of surface plasmon resonances. Proc Natl Acad Sci USA 107(33): energy in metal/dielectric structures. J Appl Phys 98(1):011101. 14530–14534.

5562 | www.pnas.org/cgi/doi/10.1073/pnas.1603536113 Tanzid et al. Downloaded by guest on October 1, 2021 13. Porto JA, Garcia-Vidal FJ, Pendry JB (1999) Transmission resonances on metallic 19. Wang Z, Bovik AC (2009) Mean squared error: Love it or leave it? A new look at signal gratings with very narrow slits. Phys Rev Lett 83(14):2845–2848. fidelity measures. IEEE Signal Process Mag 26(1):98–117. 14. Zou S, Janel N, Schatz GC (2004) Silver nanoparticle array structures that produce 20. Wang Z, Simoncelli EP, Bovik AC (2003) Multi-scale structural similarity for image – remarkably narrow plasmon lineshapes. J Chem Phys 120(23):10871–10875. quality assessment. Conf Rec Asilomar Conf Signals Syst Comput 2:9 13. 15. Kravets VG, Schedin F, Grigorenko AN (2008) Extremely narrow plasmon resonances 21. Paciaroni M, Linne M (2004) Single-shot, two-dimensional ballistic imaging through scattering media. Appl Opt 43(26):5100–5109. based on diffraction coupling of localized plasmons in arrays of metallic nano- 22. Artal P, Navarro R (1994) Monochromatic modulation transfer function of the human particles. Phys Rev Lett 101(8):087403. eye for different pupil diameters: An analytical expression. J Opt Soc Am A Opt Image 16. Chu Y, Schonbrun E, Yang T, Crozier KB (2008) Experimental observation of narrow Sci Vis 11(1):246–249. surface plasmon resonances in gold nanoparticle arrays. Appl Phys Lett 93(18): 23. Johnson PB, Christy RW (1972) Optical constants of the noble metals. Phys Rev B 6(12): – 2006 2009. 4370–4379. 17. Wang Z, Bovik AC, Sheikh HR, Simoncelli EP (2004) Image quality assessment: From 24. Yu N, et al. (2011) Light propagation with phase discontinuities: Generalized laws of error visibility to structural similarity. IEEE Trans Image Process 13(4):600–612. reflection and refraction. Science 334(6054):333–337. 18. Smith FD (1963) Optical image evaluation and the transfer function. Appl Opt 2(4): 25. Lin D, Fan P, Hasman E, Brongersma ML (2014) Dielectric gradient metasurface 335–350. optical elements. Science 345(6194):298–302. ENGINEERING

Tanzid et al. PNAS | May 17, 2016 | vol. 113 | no. 20 | 5563 Downloaded by guest on October 1, 2021