Converting RGB image to data values using existing palette in FME

Question

I have image (RGBA) data that needs to convert values ​​using an existing palette. I have the palette available to get each RGB(A) -value translated to a correct value.

I need to convert RGB value into one band. After that I need to use palette and convert each RGB value to palette value.

Example I have red color pixel (RGB 255, 0,0) in original image. Then I have value 12 in palette which has correct value to convert red pixel to value 12.

How to do that?

The palette has total 25 values and looks like this:

RGB24
1 97,100,101
2 148,147,148
3 195,204,209
4 166,159,255
5 17,37,250
6 52,92,255
8 6,231,255
10 54,199,6
12 255,0,0
.
.

I have tried the followed FME process but its wrong.

Which transformers should be used in FME?

Answer 1

So the main issue is that the RasterBandInterpretationCoercer transformer converts RGB values into uint8 values. Those values can be from 0-255, and yet you've only got 25 entries in the palette.

The other issue is that the RasterBandInterpretationCoercer transformer converts 3 RGB bands into 3 more uint8 values. I don't really know what colour you'd get if you have 3 bands all with a palette applied.

Either way, I think that a RasterExpressionEvaluator transformer will help. For example, here I use it to turn the rgb into a single uint value:

I just averaged the three RGB values into one uint8.

Now, because you've only got 25 entries in your palette, I suspect you should group cells into groups of 10. For example: 0-9 = 0, 10-19 = 1, 20-29 = 2, and so on. You could instead expand the palette to 255 values, but I don't think that's such a great method.

The easiest way to do that is to divide by 30 instead of 3 in the RasterExpressionEvaluator:

(A[0]+A[1]+A[2])/30

I tried this and it seemed to work fine. Of course, it depends on how you want to map the source to the destination cells, but if you've got 25 palette entries then they'll need the source cells grouping in some way and this seems the best to me.