[GIS] Interpolating parts of raster layer in QGIS

interpolationqgisraster

I have the following raster layer

The part in black have a value of 255, while the yellow parts have other values that goes from 0 to 255 depending on the color. The black part are the one that i want to interpolate since they have a value of 255 because no value were present, so it is like they are null values.

Is there a way to interpolate/assign a value to the black parts and obtain a raster layer without them?

Best Answer

Given your value 255 is not set as nodata ("it is like they are null values"), I would suggest two steps - (1) Assign 255 as your Nodata cell. (2) Fill Nodata cells.

(1) Set Nodata cell by gdal_translate

From QGIS Processing Toolbox, activate GDAL/OGR | [GDAL] Conversion | Translate (convert format).
There is an entry: Nodata value, leave blank to take the nodata value from input. Give it 255.
Expand Advanced parameters (small black triangle), where you will see Output raster type. Select Byte.
Click on Run. You will get a new layer Converted.

(2) Fill holes by Fill Nodata cells

From QGIS menu, open Raster | Analysis | Fill nodata. Select Converted as the Input layer.
Set Output file name to be the output and click on Run.

Footnote about the choice of tool for (1) handling nodata value:

GRASS has r.null tool which specifically handles nodata value. Today I had difficulty using this tool as it changed grid size for some unknown reason. (But it may worth a try).
SAGA also handles nodata through its Reclassify values tool, but the output becomes Float32 grid (SAGA default setting).
So I assumed gdal_translate is the tool to change nodata value for Byte-type dataset.

Related Solutions

[GIS] Extract all unique values from a RGB raster

This does not probably work generally but it should give a correct result in your case.

I captured your image above and saved it as "pngtest.png". I checked that it is a RGB png file with alpha channel. However, because it looks like a classified image I decided to try what happens if I convert it into a paletted tiff with GDAL tool rgb2pct.py http://gdal.org/rgb2pct.html

python rgb2pct.py pngtest.png palettetest.png

Then I checked what I get with gdalinfo. Colours 0-91 should be the distinct RGB values of your image, all the rest are 0,0,0,255.

gdalinfo palettetest.png
Driver: GTiff/GeoTIFF
Files: palettetest.png
Size is 1036, 731
Coordinate System is `'
Image Structure Metadata:
  INTERLEAVE=BAND
Corner Coordinates:
Upper Left  (    0.0,    0.0)
Lower Left  (    0.0,  731.0)
Upper Right ( 1036.0,    0.0)
Lower Right ( 1036.0,  731.0)
Center      (  518.0,  365.5)
Band 1 Block=1036x7 Type=Byte, ColorInterp=Palette
  Color Table (RGB with 256 entries)
    0: 48,48,48,255
    1: 104,104,104,255
    2: 40,40,40,255
    3: 96,96,96,255
    4: 144,144,144,255
    5: 64,64,64,255
    6: 16,16,16,255
    7: 200,200,200,255
    8: 136,136,136,255
    9: 56,56,56,255
   10: 8,8,8,255
   11: 0,0,0,255
   12: 216,216,216,255
   13: 32,32,32,255
   14: 248,40,0,255
   15: 24,24,24,255
   16: 176,176,176,255
   17: 248,8,0,255
   18: 168,168,168,255
   19: 248,88,0,255
   20: 248,32,0,255
   21: 248,112,0,255
   22: 248,136,0,255
   23: 248,56,0,255
   24: 248,24,0,255
   25: 248,80,0,255
   26: 248,176,0,255
   27: 248,104,0,255
   28: 248,152,0,255
   29: 248,128,0,255
   30: 248,72,0,255
   31: 248,208,0,255
   32: 248,168,0,255
   33: 248,16,0,255
   34: 248,200,0,255
   35: 248,64,0,255
   36: 248,232,0,255
   37: 216,248,0,255
   38: 248,120,0,255
   39: 248,160,0,255
   40: 232,248,0,255
   41: 248,192,0,255
   42: 184,248,0,255
   43: 248,224,0,255
   44: 248,48,0,255
   45: 200,248,0,255
   46: 232,232,232,255
   47: 176,248,0,255
   48: 152,248,0,255
   49: 224,248,0,255
   50: 248,144,0,255
   51: 248,216,0,255
   52: 120,248,0,255
   53: 144,248,0,255
   54: 104,248,0,255
   55: 168,248,0,255
   56: 208,248,0,255
   57: 136,248,0,255
   58: 248,0,0,255
   59: 72,248,0,255
   60: 248,184,0,255
   61: 96,248,0,255
   62: 48,248,0,255
   63: 128,248,0,255
   64: 88,248,0,255
   65: 192,248,0,255
   66: 248,96,0,255
   67: 8,248,0,255
   68: 40,248,0,255
   69: 80,248,0,255
   70: 0,240,8,255
   71: 32,248,0,255
   72: 160,248,0,255
   73: 0,232,16,255
   74: 24,248,0,255
   75: 112,248,0,255
   76: 0,216,32,255
   77: 0,168,80,255
   78: 0,200,48,255
   79: 64,248,0,255
   80: 240,248,0,255
   81: 112,112,112,255
   82: 0,128,120,255
   83: 0,184,64,255
   84: 0,152,96,255
   85: 0,104,144,255
   86: 0,248,0,255
   87: 0,72,176,255
   88: 248,248,0,255
   89: 248,240,0,255
   90: 240,240,240,255
   91: 248,248,248,255
   92: 0,0,0,255
   93: 0,0,0,255

[GIS] Interpolating raster to get finer resolution

If you want to merely, "Like, if there were some way to break down the pixels into small areas, but the pixels still having the same values, it also would work." Then merely change the cell size to the new number (say from 9 to 3). You can do this in QGIS via the raster calculator or gdal warp (How to resample GeoTIFF images to the same resolution?).

Now if you want to decrease the cell size you could certainly convert cells to points and then you could use kriging if required or items such as spline, weighted nearest neighbor, and IDW, but I think a simple TIN would suffice and then convert the TIN to raster. You may even be able to create the TIN direct from the raster.

This may a suitable case for Tobler's pycnophylactic interpolation in raster. This mass preserving analysis will assign values to your new cells based on their neighbors. I have only ever done this in ArcGIS but it can be run GRASS 7 using v.surf.mass. You cells would break into say 9 cells per cell and then use pycnophylactic interpolation at this level.

I would suggest you start with a TIN.

Changing Raster Resolution Resizing image resolution from 1000 m to 250 m in ArcGIS Desktop?

How to create a TIN in QGIS https://docs.qgis.org/2.6/en/docs/user_manual/plugins/plugins_interpolation.html

Best Answer

Related Solutions

[GIS] Extract all unique values from a RGB raster

[GIS] Interpolating raster to get finer resolution

Related Question