Properly configure image segmentation algorithm

OTB Usage
1

Context

I’m searching a way to properly configure the segmentation algorithm.

Configuration setup

System: Windows10
Version of the OTB: 8.0

Hello everyone !

This question has been asked 2 months ago on gis.stackexchange (link : https://gis.stackexchange.com/questions/454564/properly-configure-image-segmentation-on-qgis-segmentation-meanshift:

I’m trying to perform remote sensing on georeferenced historical satellite images (taken from Google Earth), using the segmentation algorithm from OTB.

In the parameters of the Segmentation algorithm, depending on what spatial radius and range radius I specify, the result will vary.

Is there a way to know, according to the image characteristics, which values are required to execute the best segmentation?

I have read that I could execute the ComputeImageStatistics algorithm, to get the information I need.

For my current image, the execution of this algorithm give me this result:

    <?xml version="1.0" ?>
<FeatureStatistics>
    <Statistic name="mean">
        <StatisticVector value="30.3281" />
        <StatisticVector value="40.098" />
        <StatisticVector value="38.0882" />
    </Statistic>
    <Statistic name="min">
        <StatisticVector value="0" />
        <StatisticVector value="0" />
        <StatisticVector value="0" />
    </Statistic>
    <Statistic name="max">
        <StatisticVector value="255" />
        <StatisticVector value="255" />
        <StatisticVector value="255" />
    </Statistic>
    <Statistic name="stddev">
        <StatisticVector value="30.9288" />
        <StatisticVector value="40.2171" />
        <StatisticVector value="36.3203" />
    </Statistic>
</FeatureStatistics>

But I don’t know how to analyze it. :pleading_face:

If you click on the link I gave, you’ll access the same topic, with pictures.
But to summarize, I’m trying to have the most detailed delineation of the mangrove, in a specific area.

For the same image, I did 21 attempts, with some changes on spatial radius and range radius.
(and also on options like “Parameters optimization” or “Probablility estimation” for which I don’t really know how it affects the result).

But anyway, in the end, even if the result is actually impressive, it is not always perfect (I guess we can’t have a perfect result ?). In my situation, I’m trying to compare mangrove cover between years, and any delineation error has an impact on the calculation.

Thank you for your attention,

Have a good day !