Enhanced phase correlation for image registration

Patent metadata

Field	Value
Publication number	US-2016217577-A1
Application number	US-201615004029-A
Country	US
Kind code	A1
Filing date	Jan 22, 2016
Priority date	Jan 22, 2015
Publication date	Jul 28, 2016
Grant date	—

Abstract

An image registration system and method for matching images having fundamentally different characteristics. One exemplary feature of the system and method includes the use of an enhanced phase correlation method combined with a coarse sensor model to hypothesize and match a custom match metric to determine a best solution. The system and method may be operated on a non-transitory computer-readable medium storing a plurality of instructions which when executed by one or more processors causes the one or more processors to perform the image registration method utilizing the enhanced phase correlation.

First claim

What is claimed:

1 . A non-transitory computer-readable medium storing a plurality of instructions which when executed by one or more processors causes the one or more processors to perform a method for image registration utilizing an enhanced phase correlation, the plurality of instructions for the enhanced phase correlation (EPC) comprising:
instructions for smoothing correlation output between a test image and a reference image by multiplying an amplitude shaping function with an absolute amplitude of the test image; and a correlation output representation of Y-Direction shift and X-Direction shift between the test image and the reference image, wherein a surface of the correlation output includes a correlation peak.

2 . The non-transitory computer-readable medium of claim 1 , wherein the plurality of instructions for the enhanced phase correlation further comprises:
instructions for remapping image data to create a hypothesized projection.

3 . The non-transitory computer-readable medium of claim 2 , wherein remapping the image data is accomplished by a coarse sensor model; wherein the coarse sensor model is not an exact sensor model.

4 . The non-transitory computer-readable medium of claim 3 , wherein the coarse sensor model is a pinhole camera free of any lens; and wherein the reference image is captured from an electro-optical (EO) camera.

5 . The non-transitory computer-readable medium of claim 2 , wherein remapping the image data transforms the image data to appear as if the test image was taken from a direct overhead camera.

6 . The non-transitory computer-readable medium of claim 2 , wherein the plurality of instructions further comprises instructions for performing one EPC operation and then checking an EPC correlation surface for a valid correlation peak.

7 . The non-transitory computer-readable medium of claim 6 , wherein the instructions for performing one EPC operation and then checking the EPC correlation surface for the valid correlation peak, further comprises instructions for determining a correct match between the test image and the reference image with a verification metric (VM).

8 . The non-transitory computer-readable medium of claim 7 , wherein the VM measures signal-to-noise ratio of a peak of the EPC correlation surface to area surrounding the peak.

9 . The non-transitory computer-readable medium of claim 8 , wherein the VM determines the shape of the peak of the EPC correlation surface is a uni-modal circular peak.

10 . The non-transitory computer-readable medium of claim 9 wherein the plurality of instructions further comprises an instruction for establishing a threshold and making an image registration verity decision, wherein the VM uses a Z-score normalization of a maximum phase correlation value.

11 . The non-transitory computer-readable medium of claim 10 wherein the plurality of instructions further comprises an instruction for establishing a threshold, wherein if the Z-score normalization of a maximum phase correlation is above the threshold then the image registration is verified as correct.

12 . The non-transitory computer-readable medium of claim 6 , wherein the plurality of instructions further comprises instructions for looping the one EPC operation over a limited range of scale and rotation.

13 . The non-transitory computer-readable medium of claim 12 , wherein the plurality of instructions further comprises instructions for establishing a hierarchical search pattern to increase loop rate.

14 . The non-transitory computer-readable medium of claim 13 , wherein the looping is performed on parallel threads.

15 . A phase correlation method for image registration including the steps of mapping spatial domain translations to frequency domain linear functions by first obtaining a reference image and a test image which is a translation of the reference image (Equation 1), then, denoting a Fourier transform operator (Equation 2), wherein a complex exponential factor represents a two-dimensional linear phase function corresponding to a spatial domain delta function (Equation 3), and then, obtaining a correlation output between the reference image and the test image in the frequency domain (Equation 4), wherein the improvement comprises the step of:
smoothing the correlation output by multiplying an amplitude shaping function with the test image amplitude (Equation 7 and Equation 8); and verifying a match between the test image and the reference image.

16 . The phase correlation method of claim 15 , wherein the improvement further comprises solving translation errors for both frame-to-frame and frame-to-reference video images taken.

17 . The phase correlation method of claim 15 , wherein the improvement further comprises solving misregistration errors when combined with a sensor parameter search paradigm.

18 . The phase correlation method of claim 15 , in combination with shape-based object recognition, the combination comprising the step of:
recognizing a target automatically through parametric 3D models using common geometry models for multiple modalities.

19 . The phase correlation method of claim 15 , in combination with shape-based object recognition, the combination comprising the steps of:
finding a site location in a broad area search of panchromatic satellite imagery; and locating a target within 1 meter satellite imagery using the improvement.

20 . A method of enhanced phase correlation image registration comprising the steps of:
capturing a reference image from camera on a vehicle; capturing a test image from the camera; overlaying the test image and the reference image in a processor in a computer; registering the test image and the reference image with phase correlation techniques, wherein the phase correlation techniques includes:
I T ( x, y )= I R ( x+Δx, y+Δy )   (Equation 1)

let I T (x, y), I R (x, y) represent the test image and reference image, respectively, and the test image (I T ) is a translation of the reference image (I R ); and
denoting a Fourier transform operator by ℑ, wherein Fourier shift theorem, provides:
ℑ{ I T }(ω x , ω y )= e i(ω x Δx+ω y Δy) ℑ{I R }(ω x , ω y )   (Equation 2)

wherein the complex exponential factor in Equation (2) represents a two-dimensional linear phase function, and in a spatial domain, the two-dimensional linear phase function corresponds to a delta function:
ℑ −1 {e i(ω x Δx+ω y Δy) }=δ( x+Δx, y+Δy )   (Equation 3)

the phase correlation technique correlates pre-whitened versions of the test image (I T ) and reference image (I R ) and in a frequency domain, the correlation output (denoted C Φ ) takes the following form:

C
Φ



{

I
T

,

I
R

}



Assignees

• Bae Sys Inf & Elect Sys Integ

Inventors

• Tom Victor T
• Delmarco Stephen P
• Webb Helen F

Classifications

• G06T7/32 Primary

  using correlation-based methods · CPC title

• G06V10/42

  Global feature extraction by analysis of the whole pattern, e.g. using frequency domain transformations or autocorrelation · CPC title

• G06T2200/28

  involving image processing hardware · CPC title

• G06K9/52

  Physics · mapped topic

• G06T2207/20048

  Transform domain processing · CPC title