Raster erode/dilate

FME does not offer those options out of the box but depending on what you want to do the RasterExpressionEvaluator (for example) could be useful. I doubt it'll be able to do erode and dilate though.

Pinging @dmitribagh as he might have some suggestions for you.

Sadly FME does not offer any focal analysis tools for raster. I use the command line of ImageMagick

What I do is write a temporary raster (FeatureWriter), use the SystemCaller to run the ImageMagick command(s) and then read the processed file back in with a FeatureReader.

Thanks jdh for your answer. As I am controlling the whole process with Python scripts I was about to make a workaround using OpenCV python library.

However, I will take a look at ImageMagick if the upper library is not fast enough.

Concerning the georeferencing, it could be a problem in case the erosion and/or dilation add pixel at the top or left borders, thus changing the world file. I am now not clear about how the dilating algorithm works in openCV and maybe one can specify to keep number of row and column.

Is the Orfeo Toolbox a ImageMagick plugin/derivate ?

It's an interesting challenge. If the raster is a binary image, it may not be so difficult to define a workflow to perform eroding/dilating. The attachment is a prototype with 8-connectivity: raster-erode-dilate-prototype.zip

Left: Erode <-- Center: Original --> Right: Dilate

I don't see an enhancement request in our database specifically for erosion and dilation functionality. But I do see this:

Would that help in this case? If so I can add your request there (it's already set to the top priority level, so I can't increase it any further!).

Failing that, I don't know how it could be done right now - unless you convert from raster to vector and use a Bufferer tool there - so I'll ask @dmitribagh to chime in.

As a work around: a binary raster dilate can be achieved by using (4) Affiners to create copies of the original image offset one pixel to the cardinal directions. The RasterMosaicker can be used to merge the resulting dilated image. Erode can be achieved by switching the data/no-data using a RasterExpressionEvaluator and then dilating the no-data region. Note be sure to set the image no-data value to the off-bit. if you are dealing with terrains an not binary images, you might try to stack the Affined copies into new bands of the original image and then use the RasterExpressionEvaluator to do raster math... min,max,average etc..