Module: registration
¶
Coarse to fine optical flow estimator. 

Coarse to fine optical flow estimator. 

Efficient subpixel image translation registration by crosscorrelation. 
optical_flow_ilk¶

skimage.registration.
optical_flow_ilk
(reference_image, moving_image, *, radius=7, num_warp=10, gaussian=False, prefilter=False, dtype=<class 'numpy.float32'>)[source]¶ Coarse to fine optical flow estimator.
The iterative LucasKanade (iLK) solver is applied at each level of the image pyramid. iLK [1] is a fast and robust alternative to TVL1 algorithm although less accurate for rendering flat surfaces and object boundaries (see [2]).
 Parameters
 reference_imagendarray, shape (M, N[, P[, …]])
The first gray scale image of the sequence.
 moving_imagendarray, shape (M, N[, P[, …]])
The second gray scale image of the sequence.
 radiusint, optional
Radius of the window considered around each pixel.
 num_warpint, optional
Number of times moving_image is warped.
 gaussianbool, optional
If True, a Gaussian kernel is used for the local integration. Otherwise, a uniform kernel is used.
 prefilterbool, optional
Whether to prefilter the estimated optical flow before each image warp. When True, a median filter with window size 3 along each axis is applied. This helps to remove potential outliers.
 dtypedtype, optional
Output data type: must be floating point. Single precision provides good results and saves memory usage and computation time compared to double precision.
 Returns
 flowndarray, shape ((reference_image.ndim, M, N[, P[, …]])
The estimated optical flow components for each axis.
Notes
The implemented algorithm is described in Table2 of [1].
Color images are not supported.
References
 1(1,2)
Le Besnerais, G., & Champagnat, F. (2005, September). Dense optical flow by iterative local window registration. In IEEE International Conference on Image Processing 2005 (Vol. 1, pp. I137). IEEE. DOI:10.1109/ICIP.2005.1529706
 2
Plyer, A., Le Besnerais, G., & Champagnat, F. (2016). Massively parallel Lucas Kanade optical flow for realtime video processing applications. Journal of RealTime Image Processing, 11(4), 713730. DOI:10.1007/s1155401404230
Examples
>>> from skimage.color import rgb2gray >>> from skimage.data import stereo_motorcycle >>> from skimage.registration import optical_flow_ilk >>> reference_image, moving_image, disp = stereo_motorcycle() >>> #  Convert the images to gray level: color is not supported. >>> reference_image = rgb2gray(reference_image) >>> moving_image = rgb2gray(moving_image) >>> flow = optical_flow_ilk(moving_image, reference_image)
Examples using skimage.registration.optical_flow_ilk
¶
optical_flow_tvl1¶

skimage.registration.
optical_flow_tvl1
(reference_image, moving_image, *, attachment=15, tightness=0.3, num_warp=5, num_iter=10, tol=0.0001, prefilter=False, dtype=<class 'numpy.float32'>)[source]¶ Coarse to fine optical flow estimator.
The TVL1 solver is applied at each level of the image pyramid. TVL1 is a popular algorithm for optical flow estimation introduced by Zack et al. [1], improved in [2] and detailed in [3].
 Parameters
 reference_imagendarray, shape (M, N[, P[, …]])
The first gray scale image of the sequence.
 moving_imagendarray, shape (M, N[, P[, …]])
The second gray scale image of the sequence.
 attachmentfloat, optional
Attachment parameter (\(\lambda\) in [1]). The smaller this parameter is, the smoother the returned result will be.
 tightnessfloat, optional
Tightness parameter (\(\tau\) in [1]). It should have a small value in order to maintain attachement and regularization parts in correspondence.
 num_warpint, optional
Number of times image1 is warped.
 num_iterint, optional
Number of fixed point iteration.
 tolfloat, optional
Tolerance used as stopping criterion based on the L² distance between two consecutive values of (u, v).
 prefilterbool, optional
Whether to prefilter the estimated optical flow before each image warp. When True, a median filter with window size 3 along each axis is applied. This helps to remove potential outliers.
 dtypedtype, optional
Output data type: must be floating point. Single precision provides good results and saves memory usage and computation time compared to double precision.
 Returns
 flowndarray, shape ((image0.ndim, M, N[, P[, …]])
The estimated optical flow components for each axis.
Notes
Color images are not supported.
References
 1(1,2,3)
Zach, C., Pock, T., & Bischof, H. (2007, September). A duality based approach for realtime TVL 1 optical flow. In Joint pattern recognition symposium (pp. 214223). Springer, Berlin, Heidelberg. DOI:10.1007/9783540749363_22
 2
Wedel, A., Pock, T., Zach, C., Bischof, H., & Cremers, D. (2009). An improved algorithm for TVL 1 optical flow. In Statistical and geometrical approaches to visual motion analysis (pp. 2345). Springer, Berlin, Heidelberg. DOI:10.1007/9783642030611_2
 3
Pérez, J. S., MeinhardtLlopis, E., & Facciolo, G. (2013). TVL1 optical flow estimation. Image Processing On Line, 2013, 137150. DOI:10.5201/ipol.2013.26
Examples
>>> from skimage.color import rgb2gray >>> from skimage.data import stereo_motorcycle >>> from skimage.registration import optical_flow_tvl1 >>> image0, image1, disp = stereo_motorcycle() >>> #  Convert the images to gray level: color is not supported. >>> image0 = rgb2gray(image0) >>> image1 = rgb2gray(image1) >>> flow = optical_flow_tvl1(image1, image0)
Examples using skimage.registration.optical_flow_tvl1
¶
phase_cross_correlation¶

skimage.registration.
phase_cross_correlation
(reference_image, moving_image, *, upsample_factor=1, space='real', return_error=True, reference_mask=None, moving_mask=None, overlap_ratio=0.3, normalization='phase')[source]¶ Efficient subpixel image translation registration by crosscorrelation.
This code gives the same precision as the FFT upsampled crosscorrelation in a fraction of the computation time and with reduced memory requirements. It obtains an initial estimate of the crosscorrelation peak by an FFT and then refines the shift estimation by upsampling the DFT only in a small neighborhood of that estimate by means of a matrixmultiply DFT [1].
 Parameters
 reference_imagearray
Reference image.
 moving_imagearray
Image to register. Must be same dimensionality as
reference_image
. upsample_factorint, optional
Upsampling factor. Images will be registered to within
1 / upsample_factor
of a pixel. For exampleupsample_factor == 20
means the images will be registered within 1/20th of a pixel. Default is 1 (no upsampling). Not used if any ofreference_mask
ormoving_mask
is not None. spacestring, one of “real” or “fourier”, optional
Defines how the algorithm interprets input data. “real” means data will be FFT’d to compute the correlation, while “fourier” data will bypass FFT of input data. Case insensitive. Not used if any of
reference_mask
ormoving_mask
is not None. return_errorbool, optional
Returns error and phase difference if on, otherwise only shifts are returned. Has noeffect if any of
reference_mask
ormoving_mask
is not None. In this case only shifts is returned. reference_maskndarray
Boolean mask for
reference_image
. The mask should evaluate toTrue
(or 1) on valid pixels.reference_mask
should have the same shape asreference_image
. moving_maskndarray or None, optional
Boolean mask for
moving_image
. The mask should evaluate toTrue
(or 1) on valid pixels.moving_mask
should have the same shape asmoving_image
. IfNone
,reference_mask
will be used. overlap_ratiofloat, optional
Minimum allowed overlap ratio between images. The correlation for translations corresponding with an overlap ratio lower than this threshold will be ignored. A lower overlap_ratio leads to smaller maximum translation, while a higher overlap_ratio leads to greater robustness against spurious matches due to small overlap between masked images. Used only if one of
reference_mask
ormoving_mask
is not None. normalization{“phase”, None}
The type of normalization to apply to the crosscorrelation. This parameter is unused when masks (reference_mask and moving_mask) are supplied.
 Returns
 shiftsndarray
Shift vector (in pixels) required to register
moving_image
withreference_image
. Axis ordering is consistent with numpy (e.g. Z, Y, X) errorfloat
Translation invariant normalized RMS error between
reference_image
andmoving_image
. phasedifffloat
Global phase difference between the two images (should be zero if images are nonnegative).
Notes
The use of crosscorrelation to estimate image translation has a long history dating back to at least [2]. The “phase correlation” method (selected by
normalization="phase"
) was first proposed in [3]. Publications [1] and [2] use an unnormalized crosscorrelation (normalization=None
). Which form of normalization is better is applicationdependent. For example, the phase correlation method works well in registering images under different illumination, but is not very robust to noise. In a high noise scenario, the unnormalized method may be preferable.When masks are provided, a masked normalized crosscorrelation algorithm is used [5], [6].
References
 1(1,2)
Manuel GuizarSicairos, Samuel T. Thurman, and James R. Fienup, “Efficient subpixel image registration algorithms,” Optics Letters 33, 156158 (2008). DOI:10.1364/OL.33.000156
 2(1,2)
P. Anuta, Spatial registration of multispectral and multitemporal digital imagery using fast Fourier transform techniques, IEEE Trans. Geosci. Electron., vol. 8, no. 4, pp. 353–368, Oct. 1970. DOI:10.1109/TGE.1970.271435.
 3
C. D. Kuglin D. C. Hines. The phase correlation image alignment method, Proceeding of IEEE International Conference on Cybernetics and Society, pp. 163165, New York, NY, USA, 1975, pp. 163–165.
 4
James R. Fienup, “Invariant error metrics for image reconstruction” Optics Letters 36, 83528357 (1997). DOI:10.1364/AO.36.008352
 5
Dirk Padfield. Masked Object Registration in the Fourier Domain. IEEE Transactions on Image Processing, vol. 21(5), pp. 27062718 (2012). DOI:10.1109/TIP.2011.2181402
 6
D. Padfield. “Masked FFT registration”. In Proc. Computer Vision and Pattern Recognition, pp. 29182925 (2010). DOI:10.1109/CVPR.2010.5540032