Skip to contents

Pseudo-Invariant Features based Image Matching

Source: R/pifMatch.R
pifMatch.Rd

Match one scene to another based on linear regression of pseudo-invariant features (PIF).

Usage

pifMatch(
  img,
  ref,
  method = "cor",
  quantile = 0.95,
  returnPifMap = TRUE,
  returnSimMap = TRUE,
  returnModels = FALSE
)

Arguments

img

SpatRaster. Image to be adjusted.

ref

SpatRaster. Reference image.

method

Method to calculate pixel similarity. Options: euclidean distance ('ed'), spectral angle ('sam') or pearson correlation coefficient ('cor').

quantile

Numeric. Threshold quantile used to identify PIFs

returnPifMap

Logical. Return a binary raster map ot pixels which were identified as pesudo-invariant features.

returnSimMap

Logical. Return the similarity map as well

returnModels

Logical. Return the linear models along with the adjusted image.

Value

Returns a List with the adjusted image and intermediate products (if requested).

  • img: the adjusted image

  • simMap: pixel-wise similarity map (if returnSimMap = TRUE)

  • pifMap: binary map of pixels selected as pseudo-invariant features (if returnPifMap = TRUE)

  • models: list of linear models; one per layer (if returnModels = TRUE)

Details

The function consists of three main steps: First, it calculates pixel-wise similarity between the two rasters and identifies pseudo-invariant pixels based on a similarity threshold. In the second step the values of the pseudo-invariant pixels are regressed against each other in a linear model for each layer. Finally the linear models are applied to all pixels in the img, thereby matching it to the reference scene.

Pixel-wise similarity can be calculated using one of three methods: euclidean distance (method = "ed"), spectral angle ("sam") or pearsons correlation coefficient ("cor"). The threshold is defined as a similarity quantile. Setting quantile=0.95 will select all pixels with a similarity above the 95% quantile as pseudo-invariant features.

Model fitting is performed with simple linear models (lm); fitting one model per layer.

Examples

library(terra)


## Create fake example data
## In practice this would be an image from another acquisition date
lsat_b <- log(lsat)

## Run pifMatch and return similarity layer, invariant features mask and models
lsat_b_adj <- pifMatch(lsat_b, lsat, returnPifMap = TRUE,
                         returnSimMap = TRUE, returnModels = TRUE)
# \donttest{
## Pixelwise similarity
ggR(lsat_b_adj$simMap, geom_raster = TRUE)


## Pesudo invariant feature mask 
ggR(lsat_b_adj$pifMap)


## Histograms of changes
par(mfrow=c(1,3))
hist(lsat_b[[1]], main = "lsat_b")
hist(lsat[[1]], main = "reference")
hist(lsat_b_adj$img[[1]], main = "lsat_b adjusted")


## Model summary for first band
summary(lsat_b_adj$models[[1]])
#> 
#> Call:
#> lm(formula = ref ~ img, data = df)
#> 
#> Residuals:
#>     Min      1Q  Median      3Q     Max 
#> -3.3904 -0.6021  0.0163  0.7107 24.6845 
#> 
#> Coefficients:
#>              Estimate Std. Error t value Pr(>|t|)    
#> (Intercept) -324.9027     0.9321  -348.6   <2e-16 ***
#> img           92.9473     0.2184   425.5   <2e-16 ***
#> ---
#> Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
#> 
#> Residual standard error: 1.373 on 4447 degrees of freedom
#> Multiple R-squared:  0.976,	Adjusted R-squared:  0.976 
#> F-statistic: 1.811e+05 on 1 and 4447 DF,  p-value: < 2.2e-16
#> 
# }