On-the-flying smoothing for visualisation


I spend a lot of time looking at SAR images, often at very low zooming levels.

(Please correct me if I am wrong) I believe S1TBX uses pixel sampling for visualisation when needed, i.e. it selects one pixel in a block of NxN image pixels and that becomes the display pixel. N decreasing as you zoom in, eventually reaching a value of N=1.

I was wondering whether there are plans to implement on-the-flying smoothing for visualisation, meaning that more smoothing would be applied when the entire image is displayed, less as you zoom in. Like “Raster -> Filtered Band… -> Smooth and Blur” but on-the-fly. This should reduce the visual effect of the SAR speckle noise at low zooming levels.


1 Like

I don’t think your assumption is correct. If pixel sampling was used in zooming as you describe, the presence of speckle would make the image twinkle during the zoom, which is not happening. Luis can confirm how the sampling is currently implemented.


I am mostly interested in ocean images and I see the twinkling effect very often: ships appear and disappear several times as I zoom in.

Also, it is often easier to see large-scale sea patterns (currents, weather fronts…) in the small preview image that comes with the S1 products than in S1TBX when displaying the entire image (zoom all). The S1TBX view is noisier and the patterns are difficult to see. This is why I assumed that the toolbox was using pixel sampling.

As I said, this happens in many images. Just as an example:


Ok I see what you mean. Actually on-the-fly smoothing is possible already, just right-click on a band and select “Filtered band” that gives you a selection of on-the-fly filters.

I hope this helps!

OK thank you, I am aware of the “Filtered band” function but I should have expressed my question better.

By “on-the-fly smoothing” I meant that S1TBX would automatically and dynamically choose the highest smoothing factor possible (without needlessly over-smoothing) given the size of the image region that has to be displayed and given the size of the displaying window. As far as I can see the “Filtered band” function applies a fixed smoothing factor regardless of the zooming level.

For instance, to display a 20000x20000 pix S1 image in a 1000x1000 window:

  • if the entire image has to be displayed: blocks of 20x20 image pixels are turned into 1 display pixels by smoothing (maybe simple averaging)
  • if a 1000x1000 pix image region has to be displayed: no smoothing is applied

Of course this would have a negative impact on visualisation speed, but positive on visualisation quality

CSS is correct. To reduce the amount of memory used and increase the responsiveness of the visualisation, the images are subsampled in a pyramid of different zoom levels. If the we were to instead multilook the pyramids then we would need to read all pixels when zoomed out. Currently, it may be a bit difficult to view SLCs when totally zoomed out.