Histogram Matching

Overview

The eemont package extends the ee.Image class with the method matchHistogram():

ee.Image

matchHistogram(self, target, bands[, ...])

Adjust the image's histogram to match a target image.

Usage

Histogram matching performs band-wise adjustments to match the spectral response of one image to a target image. Let’s look at a few examples of how histogram matching can be used for different platform combinations in eemont.

Landsat 5 and 8

In this example, we’ll match the histograms of a Landsat 5 image to a Landsat 8 image. This can be helpful for performing consistent time series analysis using different sensors.

First, we’ll load the source image and the target image.

source = ee.Image("LANDSAT/LT05/C01/T1_SR/LT05_195028_20110208")
target = ee.Image("LANDSAT/LC08/C02/T1_L2/LC08_196027_20130714")

Warning

Images must overlap. Cloud cover or major changes in land cover may cause inaccurate results.

Now, we’ll specify which bands should be matched. We do this using a dictionary with the source bands as keys and target bands as values.

bands = {
    # Red
    "B3": "SR_B4",
    # Green
    "B2": "SR_B3",
    # Blue
    "B1": "SR_B2"
}

Note

Any bands that aren’t listed will be removed from the matched image.

Finally, we’ll call the matchHistogram method to create a new image representing the source image matched to the target image.

matched = source.matchHistogram(target, bands)

Sentinel-2 and MODIS

In this example, we’ll match the histogram of a Sentinel-2 image to a MODIS image.

source = ee.Image("COPERNICUS/S2/20180923T081641_20180923T083023_T35PQQ")
target = ee.Image("MODIS/006/MOD09A1/2018_08_05")

Specify the matching bands between images:

bands = {
    # Red
    "B4": "sur_refl_b01",
    # Green
    "B3": "sur_refl_b04",
    # Blue
    "B2": "sur_refl_b03"
}

And match the Sentinel-2 image to the MODIS image!

matched = source.match(target, bands)

You can adjust the quality and speed of histogram matching using the maxBuckets argument. Fewer buckets will run faster but produce less accurate matching. By default, 256 buckets are used.

matched = source.matchHistogram(target, bands, maxBuckets=64)

Note

maxBuckets are automatically adjusted to the nearest power of 2, so maxBuckets=50 is the same as maxBuckets=64.

Emulating Containers

Image Scaling