IRIS service specifications
Service Description
The Change Detection Analysis (IRIS) is a processing service developed by NHAZCA S.r.l. that implements image-processing algorithms for the monitoring of ground and structural changes.
The Change Detection Analysis is conceived to work with couples of optical images, as it calculates the changes occurring in the secondary image with respect to the reference one. The processor will automatically crop the images on the maximum common area and, if needed, resample one of the two images to match the Ground Sample Distance (GSD) of the other.
An optional co-registration step can then be applied to ensure the alignment of the two images with as much precision as possible, this is achieved employing a full field displacement measurement based on the Dense Inverse Search method for Optical Flow1, which results are used to align the secondary image minimizing the residual shift.
The final product is a Structural Similarity Index Measure (SSI or SSIM) map2, that graphically and numerically displays the occurred changes.
The Structural Similarity Index is defined as:
where: \(μ_x\) (\(μ_y\)), \(σ_x\) (\(σ_y\)) and \(σ_{xy}\) are, respectively, the average (weighted with a Gaussian filter), standard deviation and cross-covariance for x (y) image patch, C1 and C2 are variables which depends on the dynamic range of the pixels.
The SSIM is computed on a sliding window (patch) of fixed size and the value is assigned to the central pixel of the window. The results consist of numerical maps with a value in the range [0 – 1] assigned to each pixel of the region of interest, where 1 is the maximum SSIM (representing a perfect similarity between images) and 0 is the minimum SSIM value (representing the maximum observable change).
To create the overview product, the change maps are then converted in a coloured overlay map which can be superimposed to the original image to create an easily and immediately understandable product in a range of 0 (no changes) – 1 (maximum change). The images obtained in this way are three-band geotiff with transparency (RGBA) that can be imported in any GIS environment. The multicolor color map adopted in the creation of the RGBA is the following:
Finally, the processor makes a binarization of the SSI asset using a threshold value, between 0 and 1, defined by the user. The obtained binary product is then used by the processor for the detected change contouring and create the Contours overview product. The contours overview is a RGBA with the pre-event single band asset in gray scale and superimposed contours in yellow.
Note
This service supports only Optical EO data.
Warning
IRIS supports only Calibrated Datasets from the following EO mission: Landsat-8, Landsat-9, PlanetScope, Pleiades-1, Sentinel-2, Vision-1, Worldview-1, Worldview-2 and Worldview-3.
Workflow
The IRIS service implements the workflow depicted below.
Inputs
The service supports EO optical Data and can analyse every band of the images. Input of the IRIS service is a couple of reflectance assets from Calibrated Dataset [CD] obtained from the Optical Calibration processor. This image pair shall be made of:
-
a pre-event reflectance asset for a CBN,
-
and a post-event reflectance asset for the same CBN.
Warning
The pair of input Geophysical Assets must be from the same sensor and shall have the same CBN (e.g. red for both pre- and post-event image).
Note
To get better results from the Change Detection Analysis service it is better to use a co-registered pair having the same GSD.
Parameters
The IRIS service requires mandatory and optional parameters. All service parameters are listed in the below Table 1.
Parameter | Description | Required | Default value |
---|---|---|---|
Optical calibrated pre-event single band asset | Reference single-band geophysical asset used in the change detection analysis | YES | |
Optical calibrated post-event single band asset | Secondary single-band geophysical asset used in the change detection analysis | YES | |
Optional mask | Refference to a single band binary asset to mask iris change detection results | NO | |
Window size | Extension of the sliding window used to perform the change detection | YES | 41 |
Threshold | A threshold value, between 0 and 1, for the detected change contouring | YES | 0.4 |
Co-registration | A flag to employ Rigid or Elastic co-regitration of input assets. None to not perform the co-registration. |
YES | Rigid |
Area of Interest | A polygon representing the area of interest to be analysed in WKT format | NO |
Table 1 - Service parameters for the IRIS processor.
Pre and post event geophysical asset
The first two mandatory parameters define the input "Reference" and "Secondary" images from Optical Calibrated Datasets. Input for Optical calibrated pre-event single band asset
and Optical calibrated post-event single band asset
parameters are the path to the single-band geophysical assets from the two Calibrated Datasets. This parameter is required to specify both the reference to the Calibrated Dataset and the band (specified as CBN) to use for the analysis (e.g. red, or green, etc).
In the definition of the input geophysical single band asset the drag and drop of the single-band asset is foreseen. This is possible by dragging and dropping one of the single-band assets (CBNs) included into a Calibrated Dataset retrieved from the features in the Results panel or in the feature basket.
Hint
To consult the bands of a Calibrated Dataset, click on the Show assets
button available near the feature title. A list with all single-band assets (CBNs) included within the Calibrated Dataset will appear under the feature title.
The drag and drop of the geophysical asset provides the input dataset reference to the service in the format:
input_dataset_reference#single-band_asset
Note
This string format is the only type of input accepted.
As an example, after the drag and drop of a feature the following string will be automatically inserted as a value for the parameter:
https://catalog.disasterscharter.org//charter/cat/[chartercalibrateddataset,{callid895}]/search?format=json&uid=call895_PT01N07_966794E007_8043502022080500000000MS00_GG003002001-calibrated#nir"
Warning
Users must drag and drop the single-band asset (e.g. "nir") into both Optical calibrated pre-event single band asset
and Optical calibrated post-event single band asset
fields. The drag and drop of the Calibrated Dataset (e.g. "[CD] PLANETSCOPE PSB.SD L3B 2022-08-05 09:01:35") is not enough.
Optional mask
In this optional parameter the user can insert a binary single band asset (e.g. a cloud mask) contained within a Dataset or into a Result feature derived from another service. If specified, this asset will be used to mask iris change detection results.
Tip
To fill the parameter just drag and drop the single band asset.
Window size
The user must provide a value for the window size, which defines the size of the sliding window in pixels, this parameter can highly influence the result of the analysis. The higher this parameter is set, the more averaged the change map will be, while the smaller and the more detailed changes can be identified at the cost of potentially noisier results. This is due to the SSIM value for each pixel being computed using the information present in the whole sliding window, thus obtaining a more localized value of the index in case of a smaller window. As a rule of thumb, the dimension of the window should be set in a range between 9 and 71. This range depends on the type of changes the user wants to identify. The output SSIM maps will have the same Ground Sample Distance of the selected band.
Warning
Window size shall be in a range between 9 and 71. The inserted value must be odd.
Note
Default value is 41
.
Threshold
In this mandatory parameter the user shall provide a value between 0 and 1 to be used for the binarization of SSI and detected change contouring.
Co-registration
In this mandatory parameter the service offers the possibility to co-register the secondary single band asset to the reference one. The user can choose among an Elastic or a Rigid co-registration of input assets. In case the co-registration is not needed, the processor does a co-location of input assets over a common grid.
The co-registration parameter can be set to one of the following values:
-
None
the co-registration is not performed, -
Rigid
the full-field displacement measured is used to estimate an Affine transformation matrix which is then applied to the secondary image, -
Elastic
the full-field displacement is used to freely warp the secondary image.
Note
Default value is Rigid
.
AOI (optional)
This last parameter (optional) may define the area of interest expressed as a Well-Known Text value.
Tip
In the definition of “Area of interest as Well Known Text” it is possible to apply as AOI the drawn polygon defined with the area filter. To do so, click on the :fontawesome-solid-magic: button in the left side of the "Area of interest expressed as Well-known text" box and select the option AOI from the list. The platform will automatically fill the parameter value with the rectangular bounding box taken from the current search area in WKT format.
Outputs
The IRIS processor provides as output the following products:
-
The co-registered pair of single-band reflectance assets
-
SSI index asset
-
SSI change detection overview
-
Binary mask from thresholding of SSI
-
Detected change contours overview
The first output product is the co-registered pair of single-band reflectance assets derived within the IRIS processing. The second one is a georeferenced single band image containing the value of the SSIM assigned to each pixel of the region of interest with values between 0 (maximum observable change) and 1 (perfect similarity between images). The third one represents a georeferenced RGB representation of the SSIM map with an associated colorbar legend. The forth product is a binary map derived from the SSI index using the user defined threshold. The last product is a RGBA with the pre-event single band asset in gray scale and superimposed detected change contours in yellow.
Warning
Output single-band reflectance products are not the original input assets of the Calibrated Datasets but are altered by the co-registration in IRIS.
IRIS Product Specifications can be found in the below tables.
Attribute | Value / description |
---|---|
Long Name | Co-registered geophysical assets |
Short Name | CBN_pre, CBN_post (e.g. nir_pre and nir_post) |
Description | Co-registered pair from input single band geophysical asset |
Data Type | As per the input images (e.g. UInt16 for reflectance) |
Band | 1 |
Format | COG |
Projection | As per the input images |
Valid Range | [1 - 10000] |
Fill Value | 0 |
Attribute | Value / description |
---|---|
Long Name | SSI index |
Short Name | ssi |
Description | SSIM value assigned to every pixel of the image |
Data Type | Float 32 |
Band | 1 |
Format | COG |
Projection | As per the input images (UInt16 for reflectance assets) |
Valid Range | [0 - 1] |
Fill Value | NaN |
Attribute | Value / description |
---|---|
Long Name | SSI change detection overview |
Short Name | contours |
Description | RGBA colour composite representing the SSIM map |
Data Type | Unsigned 8-bit Integer |
Band | 4 |
Format | COG |
Projection | As per the input images |
Valid Range | [1 - 255] |
Attribute | Value / description |
---|---|
Long Name | Binary map from SSI |
Short Name | iris-change-mask |
Description | Binary change mask derived from thresholding of SSI True=255 False=0 |
Data Type | Unsigned 8-bit Integer |
Band | 1 |
Format | COG |
Projection | As per the input images |
Valid Range | [0,255] |
Fill Value | 0 |
Attribute | Value / description |
---|---|
Long Name | Detected change contours overview |
Short Name | contours |
Description | RGBA showing the pre-event single band asset in gray scale and superimposed detected change contours in yellow |
Data Type | Unsigned 8-bit Integer |
Band | 4 |
Format | COG |
Projection | As per the input images |
Valid Range | [1 - 255] |
Filter and or Vectorize IRIS binary single band asset
IRIS's binary change asset can be spatially filtered and / or converted to polygon using the FilterVectorize service.
To further post-process the iris-change-mask
single band asset by removing small isolated clusters of pixel employ the FilterVectorize service in Filter
mode by selecting a filter threshold size value.
The binary change mask can also be converted to polygons by using the FilterVectorize service in Vectorize
mode and selecting only true values DN=255 (change).
To apply both spatial filtering and vectorization on IRIS's binary change mask employ the FilterVectorize service in the Filter and Vectorize
mode.
Warning
Only the iris-change-mask
single band asset can be used in the FilterVectorize
on-demand service, being the only discrete raster produced by the IRIS service.
-
Kroeger, T., Timofte, R., Dai, D., Van Gool, L. (2016), "Fast Optical Flow Using Dense Inverse Search", In: Leibe, B., Matas, J., Sebe, N., Welling, M. (eds) Computer Vision – ECCV 2016. Lecture Notes in Computer Science, vol 9908. Springer, Cham. DOI: 10.1007/978-3-319-46493-0_29. ↩
-
Zhou, W., A. C. Bovik, H. R. Sheikh, and E. P. Simoncelli, (2004) "Image Quality Assessment: From Error Visibility to Structural Similarity." IEEE Transactions on Image Processing. Vol. 13, Issue 4, April 2004, pp. 600–612. DOI: 10.1109/TIP.2003.819861. ↩