Remote Engineered Super Resolved Imaging Zeev Zalevsky Faculty of Engineering, Bar-Ilan University, 52900 Ramat-Gan, Israel 1 Outline •Introduction •The “SW Adaptation” Process •Diffractive type Super Resolution •Geometrical type Super Resolution •Hearing with light •Conclusions 2 Outline •Introduction •The “SW Adaptation” Process •Diffractive type Super Resolution •Geometrical type Super Resolution •Hearing with light •Conclusions 3 Introduction-Diffraction Limitation What is Resolution? Resolution is finest spatial feature that an imaging system can resolve. Resolution of optical systems is restricted by diffraction (Lord Rayleigh, Abbe), by the geometry of the detector and by the noise equivalence of its pixels. 4 Introduction-Diffraction Limitation Diffraction limitation of resolution is proportional to the F number of the imaging optics. xRES MIN {X } 1.22f # 1.22 f B 5 Introduction- Geometrical Limitation Geometrical resolution is limited by the number of detector’s pixels and their size. 6 IntroductionNoise Equivalent Resolution Noise equivalent resolution is originated by the internal noises existing within each pixel of the detector (electronic noises, shot noises etc). 7 Outline •Introduction •The “SW Adaptation” Process •Diffractive type Super Resolution •Geometrical type Super Resolution •Hearing with light •Conclusions 8 SW Adaptation Process If not resolved, is it hopeless? No, if A Priori information on the object is available!!! Types of a priori information: • A single dimensional object • Polarization restricted information • Temporally restricted signal • Wavelength restricted signal • Object shape 9 SW Adaptation Process- cont. The Suggested Solution: Having a priori knowledge of the signal may lead to super resolution using an SW (space-bandwidth) adaptation process: Adapt the SW of the signal to the acceptance SW of the system 1 for all W( x, ) Wtresh SW( x, ) 0 otherwise SW(x, )dxd N 10 Outline •Introduction •The “SW Adaptation” Process •Diffractive type Super Resolution •Geometrical type Super Resolution •Hearing with light •Conclusions 11 Diff. SR- Time Multiplexing Time Multiplexing: Conversion of temporal degrees of freedom to spatial domain (diffraction) Ap. Obj. Img. ... dt G1 G2 Synchron. moving gratings The structure of the rotated grating (for 2-D S.R. effect) Recent improvements: •Automatic synchronization (one grating, transmitted twice) •2-D objects, 2-D gratings •Dammann gratings 12 Diff. SR- Time Multiplexing Diff. SR- Time Multiplexing With clear aperture Without time multiplexing With time multiplexing 14 Remote Diff. SR Open aperture Reconstruction Projected grating Closed aperture 15 Diff. SR- Speckle Projection Closed aperture Reconstruction Coherent Incoherent 16 Remote Diff. SR- via Background Open aperture Closed aperture Reconstruction 17 Remote Diff. SR- via background, cont. Open aperture Background Reconstruction Closed aperture Closed aperture sequence Reconstruction 18 Remote Diff. SR- from satellite Numerical simulations Experimental results 19 Remote Diff. SR- via rain/droplets Open aperture Closed aperture Reconstruction 20 Spatial DLP based SR CAMERA DMD1 Two possible experimental setups. DMD2 FOV FOV SR. 21 FOV improvement 3X3 Raw images (different DMD positions) y o1 Sensor size image 20 40 60 80 y o2 20 40 60 80 20 40 60 80 100 Sensor FOV 20 20 40 60 80 20 40 60 80 100 y o4 10 y o3 20 40 60 80 100 y o5 y o6 30 40 20 40 60 80 50 60 70 80 90 10 20 30 40 50 60 70 80 90 100 20 40 60 80 20 40 60 80 110 20 40 60 80 100 20 40 60 80 100 y o7 20 40 60 80 20 40 60 80 20 40 60 80 100 20 40 60 80 100 y o8 y o9 20 40 60 80 20 40 60 80 100 20 40 60 80 100 22 FOV improvement 3X3 Two types of algorithms (LSQR, L1) LSQR N=8 N = 8 DMDs positions L1 N=8 Original FOV = middle square only 23 Spatial DLP based SR The experimental setup. (a). (c). (b). Experimental results: (a). Image captured with open iris. (b). Image captured with semi closed iris. (c). Reconstructed super resolved image (demonstrating optical zooming of 3x). 24 Optical SAR- I Focus plane 1 Sufficiently far distance to justify far field approximation Random phase 1 Imaging lens Detector Focus plane 2 A1 amplitude of image with scale 1 A1 exp(i1 ) Scale transformation (based on zero padding in Fourier domain) from 1 to 2 B2 exp(i2 ) Movement direction Remote scene A2 amplitude of image with scale 2 If MSE reaches minima then stop A2 exp(i2 ) Scale transformation (based on zero padding in Fourier domain) from 2 to 1 B1 exp(i1 ) Left: Schematic sketch of the proposed configuration. Right: The proposed iterative algorithm. 25 Optical SAR- II, Simulations 50 50 50 100 100 100 150 150 150 200 200 200 (a). (b). 250 50 100 150 250 250 200 (c). 250 50 100 150 200 250 50 100 150 200 250 Numerical simulations. (a). Original image. (b). Its reconstruction. (c). The type of reconstruction that is obtained when the phase is wrongly reconstructed. 350 300 250 200 150 100 50 0 0 50 100 150 200 250 300 The unwrapped phase of the Fourier of the original image and the reconstructed phase. Left: Original object. Middle: Original object Fourier transform. Right: Blurred image. 26 Optical SAR- III, Experimental results 27 Outline •Introduction •The “SW Adaptation” Process •Diffractive type Super Resolution •Geometrical type Super Resolution •Hearing with light •Conclusions 28 Geometrical SRIntermediate plane mask (a). (b). (a). High resolution ref. (b). Low resolution image without SR (a). (b). (c). Reconstruction: (a). Field of view border condition. (b). High resolution mask in the intermediate image plane. (c). Low resolution mask in the intermediate image plane. Mask+sensor are shifted for the micro scanning process 29 Geometrical SR- Improved Technique Intermediate Image Plane Object Plane (a). (b). Image Plane Moving Mask Lens CCD (a) A random mask with size of . (b) Optical configuration including the binary mask. The mask has to be in an intermediate imaging position. The mask can be moved only in one direction to get different images each time. 30 Geometrical SR- Improved Technique Improvements: •The mask is the only part being shifted (instead of mask+sensor). •The SR is achieved, without any spatial loss of information. •The reconstruction has a reduced sensitivity to noise. •Although the movement of the mask is in 1-D, the obtainable SR is 2-D. •The movement of the mask does not have to be in sub pixel steps. •The recovery time is improved (the reconstruction process each pixel can be treated separately & simultaneously). 31 Geometrical SR- Simulations vs. SR factor (a). (a). (b). (b). (d). (d). (d). Super Resolution Factor (Blocked Area 50%, VarNoise 0.001) (c). (c). (e). (e). (e). (f). (f). (g). (g). (h). (h). (i). (i). (a). Simulation results depicting the algorithm dependence on the super resolution factor. Image Size: 256256, Number of images(b).taken during the (c). process = 2×SR factor, percentage of the image covered by the random mask = 50%, Noise variance = 0.001. (a). Reference image. (d). (e). Low resolution images ((b), (d), (f), (h)) and their corresponding high resolution reconstructions ((c), (e), (g), (i)) for SR factor of 4X4, 8X8, 12X12 and 16x16 respectively. (g). (f). Dependence of the algorithm and the runtime on the super resolution factor. Number of images taken during the process = 2× SR factor, the size of the ring image is 256×256 pixels. It has been obtained by performing matrix inversions. The size of each matrix is 2×SR factor rows, and SR factor columns. 32 Geometrical SR- Experimental results Spherica l Mirror Auto Collimator Auto Collimator Detector Relay Lens Folding Mirror Apertur e Stop Manual micromete r Binary Mask The experimental setup 33 Geometrical SR- Experimental results Upper row: The left image: Central part of a low resolution image. The right image: Resulting reconstructed higher resolution image. Lower row: The cross sections. 34 Outline •Introduction •The “SW Adaptation” Process •Diffractive type Super Resolution •Geometrical type Super Resolution •Hearing with light •Conclusions 35 Opto-Phone: Hearing with Light Hearing with Light: Features •The ultimate voice recognition system compatible to “hear” human speech from any point of view (even from behind). •There is no restriction on the position of the system in regards to the position of the sound source. •Capable of hearing heart beats and knowing physical conditions without physical contact for measuring. 36 Opto-Phone: Hearing with Light Features- cont. •Works clearly in noisy surroundings and even through vacuum. •Allows separation between plurality of speakers and sounds sources. •Works through glass window. •Simple and robust system ( does not include interferometer in the detection phase). 37 Opto-Phone: Hearing with Light Imaging module Camera Sensor Invisible L aser projection Laser Any visible distance Camera Imaging lens Laser 38 Let’s listen…from 80m Cell phone Back part of neck Counting…1,2,3,4,5,6 Counting…5,6,7 X - movement 1 0.5 0 -0.5 Face (profile) Counting…5,6 -1 120 140 160 180 200 220 240 260 280 300 320 Heart beat pulse taken from a throat All recordings were done in a very noisy constriction site at distance of more than 80m. 39 Results: Detection of occluded objects I (a). (b). (c). (a). Camouflaged object. (b). Camouflage without the object. (c). The object (upper left part) and the low resolution camouflaged scenery. Spectrogram Spectrogram Spectrogram 0 0 0 2000 50 900 50 1800 100 50 12 800 100 100 1600 700 200 1200 600 200 200 250 500 300 400 600 350 300 350 400 400 400 200 400 450 200 450 100 450 250 1000 300 800 350 10 150 Frequency [Hz] 150 1400 Frequency [Hz] Frequency [Hz] 150 8 250 6 300 4 2 500 (d). 0 0.1 0.2 0.3 0.4 Time [sec] 0.5 0.6 0.7 500 (e). 500 0 0.1 0.2 0.3 0.4 Time [sec] 0.5 0.6 0.7 (f). 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 Time [sec] (d). The spectrogram of the camouflaged object with its engine turned on. (e). The spectrogram of the object with its engine turned on and without the camouflage. (f). The spectrogram of the camouflaged object without turning on its engine. 40 Results: Detection of occluded objects II Y - pos 10 (a). (b). 5 0 -5 Y - pos -10 10 5 -15 0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 Sample 0 -5 -10 -15 0 -20 0 1 1000 2 3 2000 4 3000 5 4000 6 5000 7 [sec] Sample (a). The scenario of the experiment. (b). Experimental results: upper recording is of the camouflaged subject. Lower recording is the same subject without the camouflage. 6000 41 Outline •Introduction •The “SW Adaptation” Process •Diffractive type Super Resolution •Geometrical type Super Resolution •Hearing with light •Conclusions 42 Conclusions: • • • • • Resolution of optical system is restricted by various terms. SW Adaptation process is a useful tool for designing super-resolution systems. A generalization for handling more types of resolution restrictions was introduced for large variety of applications. Examples of achieving super resolution effects were viewed. New approach for “hearing” with light was demonstrated. 43