Showing results for 
Search instead for 
Do you mean 

IHS Decorrelation Stretch (Saturate colors)

by Technical Evangelist ‎03-10-2020 12:22 PM - edited ‎03-10-2020 12:32 PM (108 Views)

Download model


Some imagery. especially aerial true-color imagery, can suffer from neighboring bands being highly correlated to each other. This then tends to group the RGB display of the data toward grey, or very muted, colors.


TrueColor aerial imagery (8-bit), original on the left with No Stretch, and Decorrelation Stretched on the right

The Intensity-Hue-Saturation (IHS) color transformation has long been used for purposes such as pan-sharpening via Intensity substitution methods. However it also lends itself to decorrelating RGB imagery by being able to isolate the Saturation component and boost that saturation element before transforming back to RGB space. This results in more saturated colors - reds are redder, blues are bluer, etc., This can help in visual interpretation of imagery and in various feature extraction and image classification processes.



In the provided model the user specifies which three bands represent data that is usually displayed as an RGB color composite. For true color aerial photography this would normally be 1,2,3, but may vary for other sensors. For example, IKONOS might use 3,2,1 or WorldView-3 might use 5,3,2. Bands representing wavelengths other than red, green and blue can be used (for example, a NIR, Red, Green combination), but it should be noted that if the bands start out de-correlated you may not see much improvement using this technique.

The user also specifies whether or not to automatically or manually boost the Saturation. If Automatic Mode is set to True then a scene-derived parameter is generated to automatically boost color saturation. If Automatic Mode is set to False then the value entered on the Manual Scalar port will be used to control Saturation. Manual Scalar values should vary between 0.0 and 1.0, with lower values causing higher saturation. A value of 1.0 would generally result in no change to color saturation. Values above 1.0 are allowed but will desaturate the colors, eventually driving the data toward directly correlated bands and a greyscale display. Preview mode can be used to vary the Manual Scalar value until an acceptable value is determined.

Data Requirements and Assumptions

The model has been designed with unsigned integer input data in mind.

The input imagery should have at least three bands.

It is assumed that the input data has NoData defined.  

Requires ERDAS IMAGINE 2020 Update 1 or higher.

Other Results: 

TrueColor aerial imagery (8-bit), original on the left with No Stretch, and Decorrelation Stretched on the right
WorldView-2 imagery (16-bit), original 5,3,2 on the left with Percent Stretch, and Decorrelation Stretched on the right