Home

Awesome

Digital Image Correlation using an FFT-approach

State of the Art, validated, and calibrated DIC tool - for 8bit, equal dim, single- and multi-channel images, with on-demand geotiff information forwarding

Quantify displacement of features (landslides, glaciers, etc.) in a series of images over time & retrieve displacement velocities and directions. Output data include CVP 2D-displacement maps, displacement vectors, as well as geo-rectified GIS-ready (.tif and .txt) output (optional).

V. T. Bickel & A. Manconi, May 22nd 2020 (v3) | ETH Zurich / MPS Goettingen

Please direct bugs/questions/feedback to: [valentin.bickel@erdw.ethz.ch / andrea.manconi@erdw.ethz.ch]

Please cite this routine as:

Bickel, V.T.; Manconi, A.; Amann, F. Quantitative Assessment of Digital Image Correlation Methods to Detect and Monitor Surface Displacements of Large Slope Instabilities. Remote Sens. 2018, 10, 865.

http://www.mdpi.com/2072-4292/10/6/865

MIT License - Copyright (c) 2018 Valentin Tertius Bickel & Andrea Manconi
<img src="https://github.com/bickelmps/DIC_FFT_ETHZ/blob/master/Figures/example3.png?raw=true">
Figure. DIC_FFT application example for a slope instability in the Swiss Alps: (a) slope instability displacement field over 19 years returned using DIC_FFT and two optical images, plotted on a hillshade. Images taken by an ebee drone (fixed wing) and from a plane. Blank areas indicate portions of the slope with displacements below DIC accuracy or without information. All runs were performed with a scan window size of 64 × 64 pixels and a spatial vector filter. (b) Interpolation of the Cuolm da Vi slope displacement field based on 23 GNSS measurements as direct validation. Results shows good agreement with the results produced using DIC_FFT, while DIC_FFT produces a much higher spatial resolution: The fast-moving section visible in the West of the area failed in 2016; using GNSS data alone, the pre-failure displacement is not visible. Modified from Bickel et al., 2018. More examples at the bottom of the readme!

Graphical Abstract

<img src="https://github.com/bickelmps/DIC_FFT_ETHZ/blob/master/Figures/graphic_abs.png?raw=true"> ________________________________________________________________________________________________________

Built-in pre-processing routines:

(1) Wallis Filter: Dynamic Contrast Enhancement of both input images; improves the quality of the FFT correlation significantly and helps to suppress noise.

(2) Sub-Pixel Co-Registration: Aligns both input images on a sub-pixel level; significantly improves the quality of the correlation and the absolute accuracy of the displacement measurement.

Built-in post-processing routines:

(1) RMSE Threshold Filter: Filters DIC output based on a RMSE threshold.

(2) Mean Filter: Filters DIC output with a algorithmic mean kernel.

(3) Median Filter: Filters DIC output with a median kernel.

(4) Vector Filter: Filters DIC output based on the pixel neighborhood's displacement vector direction & magnitude.

<img src="https://github.com/bickelmps/DIC_FFT_ETHZ/blob/master/Figures/glacier.gif?raw=true">
Figure. Example of DIC_FFT application for glacier displacement monitoring. The subplots show 1) primary and 2) secondary images, 3) the displacement magnitude, as well as 4) the displacement vectors. More examples at the bottom of the readme!

Tool is being used by:

Publications

Required Matlab toolboxes:

External/integrated functions/algorithms:

Hardware requirements:

Instructions

  1. Clone and unzip DIC_FFT_ETHZ_Repo

  2. Place primary and secondary image(s) in the "Input" folder (or copy images from 'Example' folder for a demo)

  3. Customize setup for Preprocessing, DIC, and Postprocessing in run_pixel_offset.m, using the following variables. More information about the variables can be found further below:

       Inputs ----------------------------------------------------------------------------------
        - geotiff = 0/1
        - epsg = int.
        - coreg = 0/1
        - way = 1/2
        - yourtext = 'string'
        - pix = float.
        - master = 'string'
        - slave = 'string'

   Preprocessing (Wallis filter & Co-registration) -----------------------------------------
        - wallis = 0/1
        - win = int.
        - tarm = int.
        - tars = int.
        - b = 0-1
        - c = 0-1

        - sp = int.
        - co_os = int.

   DIC -------------------------------------------------------------------------------------
        - wi = 2^n, n = int.
        - os = int.
        - pix = float./int.

   Postprocessing (RMSE, Mean, Vector, Median filter) --------------------------------------
        - filter = 1-4

        - thr = 0-1

        - mfws = [int. int.]
        - cut = int.

        - magcap = wi/int. or int.
        - xcap = wi/int. or int.
        - ycap = wi/int. or int.

        - med = [int. int.]

    Additional settings --------------------------------------------------------------------
        - scalax = [int. int.]
        - scalay = [int. int.]

        - coppia = 'string'

        - skip_x = wi/int.
        - skip_y = wi/int.

  4. Execute run_pixel_offset.m

  5. Collect displacement matrices ascii and deformation magnitude preview tif from "Output" folder for further utilization of results (e.g. drag & drop into a GIS). See "Graphical Abstract" for the organization of the output.

Input Parameter Description

Inputs - - - - - - - - - -

geotiff

indicate here, if you would like to forward geotiff information (=1) or not (=0)

epsg

in case you would like to forward geotiff information, enter the desired EPSG code here --> you can get this number here: https://spatialreference.org/ref/epsg/

coreg

indicate here, if you would like to manually select a region for the alignment (=1) or not (=0)

way

specify is you want the geotiff preview to be 8bit or 16bit

yourtext

insert a name for your output files

pix

insert the spatial resolution of the images used (GSD) in m/pixel

master (primary image)

Enter the name of the old image here, including the file type

slave (secondary image)

Enter the name of the new image here, including the file type

Preprocessing (Wallis filter & Co-registration) - - - - - - - - - -

wallis

indicate here, if you would like to apply a Wallis filter (=1) or not (=0)

win

specify Wallis filter window size, use even numbers only! A larger window will result in a larger context being considered for the dynamic contrast adaptation

tarm

specify the target mean value within each individual window - this mainly controls the image brightness (use the histogram of the filtered image to adapt the filter settings)

tars

specify the target standard deviation within each individual window - this mainly controls the radiometric normal distribution and contrast (use the histogram of the filtered image to adapt the filter settings)

b

specify the brightness enforcing constant (use the histogram of the filtered image to adapt the filter settings)

c

specify the contrast enforcing constant (use the histogram of the filtered image to adapt the filter settings)

sp

define the desired image split that is being used for the Co-Registration - 1 means that the entire image is matched, e.g. 8 means that the image is divided in 8 sub-parts, while every individual sub-image is being matched, where the mean correction of all 8 tiles will be used for Co-Registration

co_os

apply a oversampling factor for the Co-Registration, if desired - theoretically, a larger oversampling factor will result in a more accurate (sub-pixel) Co-Registration - in practice, a factor of 1 or max. 4 will be sufficient

DIC - - - - - - - - - -

wi

specify the desired window size in pixels; stick to the 'power of 2 rule' - a larger window will produce a better correlation and a better coverage of the displacement; a smaller window size helps to resolve smaller spatial scales of the displacement to the cost of a higher noise level; accuracy of the derived displacement is not influenced by the window size, see: http://www.mdpi.com/2072-4292/10/6/865

os

specify the oversampling factor of the FFT correlator - theoretically, a larger oversampling factor will result in a more accurate correlation (sub-pixel) - in practice, a factor of 1 or max. 4 will be sufficient, see http://www.mdpi.com/2072-4292/10/6/865

pix

enter the spatial resolution (GSD) of the used images in m/pixel

Postprocessing (RMSE, Mean, Vector, Median filter) - - - - - - - - - -

filter

choose whether you would like to use the RMSE filter (=1), the arithmetic mean filter (=2), the spatial vector filter (=3), or the median filter (=4) - the filters are described here: http://www.mdpi.com/2072-4292/10/6/865

thr

specify the RMSE threshold - a lower value will result in a strict filter, whereas a higher value will let pass values with a lower correlation quality (RMSE); a thr = 1 will result in a leaky filter, i.e., no filtering is performed at all - use this when you want to avoid any influence of a post-processing routine!

mfws

specify the mean filter window size in pixels - a larger window will result in a strongly blurred displacement map

cut

specify the mean filter cutoff value in pixels - this can be used to eliminate extreme noise that otherwise would mess with the mean filter

magcap

specify the magnitude (resultant) cap in pixels - values which are greater as this value are cut

xcap

specify the x direction (E-W, left-right) cap in pixels - values which are greater as this value are cut

ycap

specify the y direction (N-S, up-down) cap in pixels - values which are greater as this value are cut

med

specify the median filter window size in pixels - a larger window will result in a strongly blurred displacement map

Additional settings - - - - - - - - - -

scalax

specify the Min and Max values in x direction (E-W, left-right) that will be displayed in meters - everything above and below this range will be saturated at the given Min / Max value; in general, this range should be small for slow displacements and large for fast displacements; at the same time, regions with heterogeneous displacement velocities could require to this code with different ranges in order to catch the entire variety of displacement velocities!

scalay

specify the Min and Max values in y direction (N-S, up-down) that will be displayed in meters - everything above and below this range will be saturated at the given Min / Max value; in general, this range should be small for slow displacements and large for fast displacements; at the same time, regions with heterogeneous displacement velocities could require to this code with different ranges in order to catch the entire variety of displacement velocities!

coppia

provide a name for the ascii file that contains the displacement information (Output folder)

skip_x

specify the search window skip in x direction (E-W, left-right) - be aware of aliasing effects when going below the Nyquist limit (default)!

skip_y

specify the search window skip in y direction (N-S, up-down) - be aware of aliasing effects when going below the Nyquist limit (default)!

Selection of results

<img src="https://github.com/bickelmps/DIC_FFT_ETHZ/blob/master/Figures/example2.png?raw=true">
Figure. (a) Optical image: Surface velocity at Moosfluh, Switzerland, measured with ground-based remote sensing methods. Average velocity obtained applying DIC on time lapse camera data (image pair 26–28 September 2018) acquired from across the valley. Unreliable and/or outlier DIC values are masked out. The large deformation occurring in the lower portions of the Moosfluh slope is well mapped from both techniques in a short temporal frame, and average velocities are in agreement with independent ground measurements. (b) Ground-based Radar Interferometry measurements (stack of 1 min interferograms over 8 h) performed at Moosfluh on 28 September 2017. Results shows good agreement with the results produced using DIC_FFT (a). Modified from Manconi et al., 2018.
<img src="https://github.com/bickelmps/DIC_FFT_ETHZ/blob/master/Figures/example1.png?raw=true">
Figure. DIC_FFT applied on Sentinel-1 and Sentinel-2 datasets over the Moosfluh unstable slope. White boundary delineates the approximate area of the Aletsch glacier and Moosfluh slope instability. (a) SAR intensity image: Resultant of the surface velocities measured across-track & along-track directions of Sentinel-1 in the period 8 June–18 October 2017. Background map is SLC image 8 June 2017 (slant range oriented towards north); (b) Multispectral image: Resultant of the horizontal surface velocities (E-N directions) measured on Sentinel-2 track R108 and R065 in the periods 19 July–17 October and 26 June–14 October 2017, respectively. Background maps are the surface reflectance for the images 19 July 2017 and 26 June 2017, respectively. Surface velocities in the lower portions of the Moosfluh instability and in the central part of the Great Aletsch glacier are well mapped by DIC. Inaccurate/outlier DIC values are masked out. Modified from Manconi et al., 2018.
MIT License - Copyright (c) 2018 Valentin Tertius Bickel & Andrea Manconi