Slide 1 - UBC - Computer Science by gjjur4356

VIEWS: 3 PAGES: 14

									  Joint transform optical correlation
applied to sub-pixel image registration



 Thomas J Grycewicz, Brian E Evans, Cheryl S Lau

   Sensor Engineering & Exploitation Department
           The Aerospace Corporation
                   Chantilly, VA
Goals

Our goal is to achieve high resolution image
registration. We also want to operate at near-real-
time rates.
Basic Algorithm

Imagine flying over some terrain in an airplane
capturing video. Consecutive frames in the video
sequence contain redundant scenes. We exploit
this redundancy by taking multiple frames of the
same scene, lining them up, and adding them
together to increase the data we have about that
scene. We line the images up on a sub pixel level,
then add them together to increase resolution.
Applications

• suppress motion blur
• increase resolution
• noise cancellation (transient noise, hot and dead
pixels)
• relax requirements on focal plane
Joint Transform Correlator System


                                                       E( )
          input(x,y)                                                                 F o u rie r p la ne
                       y                                                            c am e ra

  Laser                    s
  Light                r                                 
            x                                                                              Analog
                                                                                           joint power
           Binary joint                       B inarization                                spectrum
           power
           spectrum
  Laser                                                                                    O u tp u t
  Light                                                  x                                 cam era
                                                                     y
                                   T(  )                   o(x,y)

                                                                           Illustration courtesy U.S.Air Force
Joint Transform Correlator Output
                               Power Spectrum
         Input                 (of one smiley face)




             FT

   Joint Power Spectrum            Output
                                                  correlation peaks
                          FT



                                         Illustration courtesy U.S.Air Force
Input Plane




USAF 1951 test pattern   Laplacian filtered input
 (unprocessed input)
Fourier Plane




 Fourier plane image   apply high pass filter
Output Plane




    output image   output peak
Results

With Laplacian input preprocessing, we were able
to achieve 0.246 pixel accuracy.
       ¼ pixel
  Results
             5 frames   32 frames




not
registered




registered
to ½ pixel
 All Optical System


HeNe    Spatial         BS 1
Laser   Filter          90/10




   Input SLM      HWP   BS 2    FTL 1   FP SLM    HWP BS 3 FTL 2    Output
   BNS 256x256          (POL)   200mm   Hamamatsu     (POL) 175mm   Camera
                                        OASLM
All Optical System - Results

With Laplacian input preprocessing, we were able
to achieve 0.121 pixel accuracy.
       1/8 pixel




      output image                output peak
Thank you for listening!



        Cheryl Lau

								
To top