# Creating a normal map texture using a heightmap tool

This topic is 1823 days old which is more than the 365 day threshold we allow for new replies. Please post a new topic.

## Recommended Posts

Hi, I'm trying to create a normal mapping tool that creates normalmaps from heightmaps. I've got it kind of working but there is a lot of random color and not really very smooth transitions between some pixels, there are sometimes 'criss cross' style patterns and I really have no idea whats going on.

Take a look here as gamedev.net won't allow my to img src a bmp for some reason, sorry.

http://www.flickr.com/photos/56802324@N04/8400199937/

My assumption is this: I read a 1 byte value from a height map - this can contain a value from 0 to 256. Using some samples from surrounding pixels I can try and create a 'vector', normalize this and then use the values as the RGB values of a normal map.

Here is the meat of the code:

void CreateNormalMap(const char* inFilename,const char* outFilename)
{
unsigned int width,height,bytes;

//read original bmp info
unsigned char* img = OpenImage(inFilename,width,height,bytes);

//create normal map data

//set 'scales', using a 1.0f scale for y right now,
//x and z are determined by the width/height of the bmp file
float heightScale = 1.0f;
float xScale = 1.0f/width;
float zScale = 1.0f/height;

unsigned char* normalMap = new unsigned char[width * height * 3];

for( int z = 0; z < height; z++){
for(int x = 0; x < width; x++){

int index = z * width + x;

//take sample from current pixel
float const y0 = img[index];

//take samples from surrounding pixels
float const Az = ( x + 1 < width )  ? ( img[index + 1] )     : y0;
float const Bz = ( z + 1 < height ) ? ( img[index + width])  : y0;
float const Cz = ( x - 1 >= 0 )	    ? ( img[index - 1] )     : y0;
float const Dz = ( z - 1 >= 0 )	    ? ( img[index - width] ) : y0;

//construct a vector then normalize it
glm::vec3 normVec =  glm::vec3(Cz - Az,heightScale,Dz - Bz);
normVec = glm::normalize(normVec);
//write the values to the normal map, they are currently between 0.0 and 1.0f
//so scale them using 256 to map to rgb value
normalMap[index * 3    ] = normVec.x * xScale * 256;
normalMap[index * 3 + 1] = normVec.y * yScale * 256;
normalMap[index * 3 + 2] = normVec.z * zScale * 256;
}
}

}

##### Share on other sites
How are you writing the normal map to a file? One problem that may be popping up is that your output file format is 32-bit, so that the various RGB values are not being written in the correct places (since a 32-bit format expects the values to be written RGBXRGBX... or XRGBXRGB... and you might be writing the values RGBRGBRGB)

##### Share on other sites

[quote name='Lil_Lloyd' timestamp='1358744285' post='5023765']
this can contain a value from 0 to 256
[/quote]No, it's 0-255 bounds included. That is, float(1.0) maps to ubyte(255).

Personally I have no idea what xScale and yScale are meant to do. They are the inverse of image dimensions in pixels... and they contribute to normal computation? This implies that large heightmaps are always (0,0,0), which BTW is not a valid normal!

The way you fetch neightbour pixels is debatable. What you do is actually clamping but it's way more likely you'll want wrapping instead.

There are various filters to deal with heightmaps -> normal conversion. Saving it as jpg was a really bad choice, use PNG instead.

Take a look at GIMP's normalmap plugin for the various available options, it's really more complicated than it seems.

##### Share on other sites

I tried installing GIMP'S normal map plugin after updating to the latest version but for some reason it's always greyed out in the menu. Anyway, thank you for pointing out the 0-255 gotcha, I noticed it myself a few moments before returning to this thread and fixed it, the xScale/yScale thing was really dumb, so I killed that.

A lot of the black spots from before are now missing as they were due to overflow before, and the fidelity of the bitmap file I was using definitely caused a few problems. Now I have started using a better quality .bmp and the produced normal maps are better, but still have some strange artefacts/rings, but maybe a Gaussian blur will help in a paint package.

http://www.flickr.com/photos/56802324@N04/8400575999/in/photostream