Light field microscopy Marc Levoy, Ren Ng, Andrew Adams Matthew Footer, Mark Horowitz Stanford Computer Graphics Laboratory Executive summary • captures the 4D light field inside a microscope • yields perspective flyarounds and focal stacks from a single snapshot, but at lower spatial resolution • focal stack → deconvolution microscopy → volume data Marc Levoy Devices for recording light fields (using geometrical optics) big scenes small scenes • • • • • handheld camera camera gantry array of cameras plenoptic camera light field microscope [Buehler 2001] [Stanford 2002] [Wilburn 2005] [Ng 2005] (this paper) Marc Levoy Light fields at micron scales • wave optics must be considered – diffraction limits the spatial × angular resolution • most objects are no longer opaque – each pixel is a line integral through the object » of attenuation » or emission – can reconstruct 3D structure from these integrals » tomography » 3D deconvolution Marc Levoy Conventional versus plenoptic camera Marc Levoy Conventional versus plenoptic camera 125μ square-sided microlenses uv-plane st-plane Marc Levoy Digital refocusing Σ Σ • refocusing = summing windows extracted from several microlenses Marc Levoy Example of digital refocusing Marc Levoy Refocusing portraits Marc Levoy Macrophotography Marc Levoy Digitally moving the observer Σ Σ • moving the observer = moving the window we extract from the microlenses Marc Levoy Example of moving the observer Marc Levoy Moving backward and forward Marc Levoy A light field microscope (LFM) eyepiece intermediate image plane objective specimen Marc Levoy A light field microscope (LFM) sensor eyepiece intermediate image plane objective specimen → reduced lateral resolution on specimen = 0.26μ × 12 spots = 3.1μ • 40x / 0.95NA objective ↓ 0.26μ spot on specimen × 40x = 10.4μ on sensor ↓ 2400 spots over 25mm field • 1252-micron microlenses ↓ 200 × 200 microlenses with 12 × 12 spots per microlens Marc Levoy A light field microscope (LFM) sensor 2.5mm 160mm 0.2mm Marc Levoy Example light field micrograph • • • • • • orange fluorescent crayon mercury-arc source + blue dichroic filter 16x / 0.5NA (dry) objective f/20 microlens array 65mm f/2.8 macro lens at 1:1 Canon 20D digital camera 200μ ordinary microscope light field microscope Marc Levoy The geometry of the light field in a microscope • microscopes make orthographic views • translating the stage in X or Y provides no parallax on the specimen f • out-of-plane features don’t shift position when they come into focus objective lenses are telecentric Marc Levoy Panning and focusing panning sequence focal stack Marc Levoy Mouse embryo lung (16x / 0.5NA water immersion) 200μ light field pan focal stack Marc Levoy Axial resolution (a.k.a. depth of field) • wave term + geometrical optics term DOFtot λn n e 2 NA M NA • ordinary microscope (16x/0.4NA (dry), e = 0) 0.535 1 3.3 2 0.4 (wave optics dominates) • with microlens array (e = 125μ) (geometrical optics dominates) 0.535 1 1 125 3.3 19.5 22.8 2 0.4 16 0.4 • stopped down to one pixel per microlens 3.3 19.5 12 spots 237 → number of slices in focal stack = 12 Marc Levoy 3D reconstruction • confocal scanning [Minsky 1957] • shape-from-focus [Nayar 1990] • deconvolution microscopy [Agard 1984] – 4D light field → digital refocusing → 3D focal stack → deconvolution microscopy → 3D volume data (UMIC SUNY/Stonybrook) (Noguchi) (DeltaVision) Marc Levoy 3D deconvolution [McNally 1999] focus stack of a point in 3-space is the 3D PSF of that imaging system • • • • • • object * PSF → focus stack {object} × {PSF} → {focus stack} {focus stack} {PSF} → {object} spectrum contains zeros, due to missing rays imaging noise is amplified by division by ~zeros reduce by regularization, e.g. smoothing {PSF} Marc Levoy Silkworm mouth (40x / 1.3NA oil immersion) 100μ slice of focal stack slice of volume volume rendering Marc Levoy Insect legs (16x / 0.4NA dry) 200μ volume rendering all-focus image [Agarwala 2004] Marc Levoy 3D reconstruction (revisited) • 4D light field → digital refocusing → 3D focal stack → deconvolution microscopy → 3D volume data (DeltaVision) • 4D light field → tomographic reconstruction → 3D volume data (from Kak & Slaney) Marc Levoy Implications of this equivalence • light fields of minimally scattering volumes contain only 3D worth of information, not 4D • the extra dimension serves to reduce noise, but could be re-purposed? Optical Projection Tomography [Sharpe 2002] Marc Levoy Conclusions • captures 3D structure of microscopic objects in a single snapshot, and at a single instant in time Calcium fluorescent imaging of zebrafish larvae optic tectum during changing visual stimula Marc Levoy Conclusions • captures 3D structure of microscopic objects in a single snapshot, and at a single instant in time but... • sacrifices spatial resolution to obtain control over viewpoint and focus • 3D reconstruction fails if specimen is too thick or too opaque Marc Levoy Future work • extending the field of view by correcting digitally for objective aberrations Nikon 40x 0.95NA (dry) Plan-Apo Marc Levoy Future work • extending the field of view by correcting digitally for objective aberrations • microlenses in the illumination path → an imaging microscope scatterometer 200μ angular dependence of reflection from single squid iridophore Marc Levoy http://graphics.stanford.edu/projects/lfmicroscope Marc Levoy