Colour Correction using Histogram Stretching

Colour Correction using Histogram Stretching Part of series on

Integrating space assets in airport management and operation

GLOBAL AVIATION INNOVATION

INCREASING FLIGHTSAFETY

USER FRIENDLY SOFTWARE SOLUTIONS

COMPLIANCE

ASCEND XYZ

Histogram Stretching

Scope

The challengeThe input data is not distributed in the colour spectrum.

The solution Ascend uses statistics to stretch input data to the full colour spectrum and tests hypothesis on disregarding cloud information to improve visual appeal.

Original Image clipped by region and clouds shown.

Naive histogram stretching

• Data is stretched using

mean ± 2 standard

deviations

• Performed band-wise

• Statistics are calculated

from all data

Original RGB Band (16 bit)

Results : Huge Improvements

Naive Histogram Stretching

Vector dataUsing Clouds to improve algorithm

Area of interest given byGeoJSON file and clouds given by GML.

These are then put on top of the original RGB image, to show clouds on map.

Roskilde Airport is in the middle of the map

Histogram stretching of cloud data• Data is stretched using mean

± 2 standard deviations

• It is performed band-wise

• Statistics are calculated only from areas not covered by clouds

Take it to the next level?

Improved Histogram Stretchingomitting clouds from statisticsNaïve Histogram Stretching

Improvements are seen

Original Sentinel RGB Bands Naive Histogram Stretching

Another Example

Another Example

Naive Histogram Stretching Histogram Stretching discarding cloud data for statistics

Experiment: Small non cloud area used

Naive Histogram Stretched dataCloud Mask: Green is clouds, grey is no clouds.

Example: When the visibility is better some places, and worse in other places.

Naive histogram Stretching Histogram without cloud data

Discussions

The naive histogram stretching clearly improve visual appeal for end users and has been implemented for first release.

Does the improved histogram stretching, using disregard of clouds in data statistics for the histogram stretching, improve the visual experience for the end user?

Is it worth the resources to move the improved algorithm into production?

How do we reduce the time from moving research experiments into production?

What could make the visibility even better?