US 12,462,340 B2
Device and method for correspondence analysis in images
Marc Schulze, Dresden (DE); Joachim Ihlefeld, Dresden (DE); and Torvald Riegel, Dresden (DE)
Assigned to recognitionfocus GmbH, Dresden (DE)
Appl. No. 18/275,273
Filed by recognitionfocus GmbH, Dresden (DE)
PCT Filed Jan. 31, 2022, PCT No. PCT/EP2022/052201
§ 371(c)(1), (2) Date Aug. 1, 2023,
PCT Pub. No. WO2022/162216, PCT Pub. Date Aug. 4, 2022.
Claims priority of application No. 102021102233.9 (DE), filed on Feb. 1, 2021.
Prior Publication US 2024/0303772 A1, Sep. 12, 2024
Int. Cl. G06K 9/00 (2022.01); G06T 5/20 (2006.01); G06T 7/11 (2017.01); G06T 7/593 (2017.01); G06V 10/56 (2022.01); G06V 10/764 (2022.01)
CPC G06T 5/20 (2013.01) [G06T 7/11 (2017.01); G06T 7/593 (2017.01); G06V 10/56 (2022.01); G06V 10/764 (2022.01)] 26 Claims
OG exemplary drawing
 
1. A correspondence analyzer for determining a disparity δ, that is a shift between corresponding image elements in two digital individual images, comprising:
a computing device configured:
to select image patches from the two individual images in each case, the image patch of one of the individual images being chosen as a reference image patch, and a sequence of search image patches being selected in the other individual image; and
to generate a plurality of signals YLsignal,v from the reference image patch and a plurality of signals YRsignal,v from each of the search image patches; and
to perform a convolution of the plurality of signals YLsignal,v of the reference image patch with even and odd convolution kernels stored in a memory, in a spatial window, wherein the even convolution kernels comprise a weighted sum of a plurality of even harmonic functions of different spatial frequencies and the odd convolution kernels comprise a weighted sum of a plurality of odd harmonic functions of different spatial frequencies; and
to perform a convolution of the signals YRsignal,v for each of the search image patches with the convolution kernels stored in the memory, in the spatial window; and
to calculate the differences of the respective convolution results for each signal pair YLsignal,v and YRsignal,v; and
to process the differences of the convolution results for each of the search image patches in a non-linear manner and to accumulate them to obtain a function value of a correspondence function SSD(δp) at the point δp, or to calculate, from the differences in the convolution results, the first derivative SSD(δp) of the correspondence function SSD with respect to δp at the point δp, and thus to obtain a function value of a correspondence function SSD or of its derivative at the point δp, wherein & denotes the distance of the reference image from the search image; and
to determine local extrema of the correspondence function SSD or zero crossings of the derivative SSD(δp) of the correspondence function SSD(δp); and
to output the point δp of one of the local extrema or of one of the zero crossings as the disparity δ; or
to calculate and output a subpixel-precise value of the disparity at the point δp.