Lighting Your Texture Maps Introduction So, you've got your engine running, you have texture mapping, and some kind of shading in place, it runs fast, its smooth, but you feel somethings missing. You need to shade your texture maps. But you haven't got a clue how to do it! Well, I shall now explain the simple process of adding shading or transparency to your textures, in 256 colour mode. 16*16 Linear Palette The easiest way to make primitive shading is to change your palette. Firstly, you will only be able to have 256 / n shades, where n is the number of colours you will be using. Secondly, the image quality is very poor. But, I'll explain it anyway, and you should at least implement it, if at least just to sharpen your coding skills. You'll find it leads on well to the better method... Ok, the idea here is that instead of having just one big lump of 256 colours, you organise the colours into a table. Say a 16*16 table. Going down the rows you have increasing shade, and going across the table you have colours. Building this table is easy. You start at black for the first row, for every colour. Then, you fade up to the correct RGB in each successive row. So for colour 1: R, G, B = RGB values for the colour DeltaR, DeltaG, DeltaB = Amount to step at each row DeltaR = (R << 8) / numshades DeltaG = (B << 8) / numshades DeltaB = (B << 8) / numshades CurR, CurG, CurB = current red, green, blue CurR = CurG = CurB = 0 for row = 0 to numshades - 1 for colour = 0 to numcolours - 1 palette[row*16+colour].red = CurR[colour] >> 8 palette[row*16+colour].green = CurG[colour] >> 8 palette[row*16+colour].blue = CurB[colour] >> 8 CurR[colour] += DeltaR[colour] CurG[colour] += DeltaG[colour] CurB[colour] += DeltaB[colour] end end This is only pseudo-code. Coding this up is a 15-min maximum job. OK, now you've got that going, you need to light your textures. Again, this is easy. Simply do: PutPixel(x, y, shade*16+texture[v][u]) Simple! Hold on to this idea of using a lookup table to get your colour, its a very important concept. Using A Seperate Look-up Table In the previous section, the lookup table was your palette. Now, we will create a seperate lookup table, which will let us do better shading, and improve the versatility of our code. Allocate a 256*256 block of memory. If you're still in real-mode, and can't allocate this much memory, stop for a minute, and ask yourself what the hell you're playing at? You're using a computer that has somewhere between 4 and 32 megs of RAM installed, and you're constricting yourself to 1 meg? Get yourself a new compiler! If you're a C/C++ man, get DJGPP or Watcom C++. If Pascal is your thing, go for TMT Pascal, or perhaps GNU Pascal under Linux. Finally, if you're a masochist, and use 100% assembly, get yourself DOS32, or Trans PMODE extender. Personally I prefer DOS32 and Watcom. Okay, that block of memory is going to be used for a lookup table. As before, you're going to increasing shade with increasing row, and colour running across the columns: Colour --> +----------------------+ Shade 0 | | 1 | | ... ~~~~~~~~~~~~~~~~~~~~~~ 256 | | +----------------------+ That really should have been a .gif picture, but what the hell, you can see what I mean. Now you'll need to fill that table. And this is the tricky, time consuming bit. Finding The Colour Of Best Fit We only have a finite number of colours in our palette. And to fully shade the texture, we would need 256*256 colours. So, we need to find the best fitting colour for a given shade. The way this is done is pretty simple. You need to find the 'distance' between the target colour, and the colour in the palette. This gives you a measure of how close the colour is to the desired colour. Then return the colour with the minimum distance. Distance is found by : Dist = sqrt((TargetR - R) ^ 2 + (TargetG - G) ^ 2 + (TargetB - B) ^ 2) Set your best-fit distance variable to some high, impossible number, say a million. Loop for every colour, and if dist < best-dist, then set best-dist to dist, and best-colour to current-colour. If distance is 0, then just return the colour; you can't do better than a perfect fit. Otherwise, at the end of the loop, just return your best fitting colour. Your code should look something like : int BestColour(float r, float g, float b, char *palette) { int n; int bestcol = 0; int bestdist = 256000; float dist, rdist, gdist, bdist; float red, green, blue; for(n=0; n<256; n++) { red = (float) palette[0]; green = (float) palette[1]; blue = (float) palette[2]; rdist = red - r; gdist = green - g; bdist = blue - b; dist = sqrt(rdist * rdist + gdist * gdist + bdist * bdist); if(dist <= bestdist) { bestdist = dist; bestcol = n; } if(dist == 0) return n; palette += 3; } return bestcol; } Its also a good idea to weight the rgb distances differently. Your eyes are generally more sensitive to the green band of colour, then red, then blue. So weight them perhaps: 0.30*Red 0.59*Green 0.11*Blue This usually improves quality a bit. Also, you can avoid the sqrt() function call, by just comparing the squared distance. I'm not sure if this works perfectly as a replacement, but it seems fine to me. Please note that this is a pre-process, NOT to be done at rendertime (unless palette changes) Calculating The Colour For A Given Shade Calculating what colour to search for is simple enough. You can use any formulae you like for this, a phong lighting model, a transparency, a depth cue formula. The point is you need to calculate RGB for a given colour, and a given 'shade'. Phong Lighting Model The formula for this is: Ambient + Diffuse + Specular Which becomes: Ia*ka*Oa + I*(Od*kd*(N.L) + Os*ks*(R.V)^n) Ia is intensity of ambient Ka is ambient co-ef Oa is ambient colour. These are constant, set to whatever you like I is intensity of light. Say 1.0 Od is colour for diffuse, replace with colour[n] Kd is diffuse co-efficient. About 0.95 - 1.0 is good (N.L) is angle of incidence, replace shade[n] Os is specular colour, say 255 Ks is specular co-ef, say 0.75 (R.V) is angle of reflection. Replace with spec[n] N is the shinyness. A value of around 20 looks fine. You'll need to calculate this for R G and B seperately. Make the N.L and (R.V)^n terms into lookup tables. Then search for the value in the palette, and you're done. Depth cueing A simple approximation can be done with k*Colour Where k is shade/256. Its also a good idea to make the colour fade to white, instead of black, which gives the illusion of fogging. Transparecy Transparecy is slightly different. Instead of shade, you will have background-colour, and colour will be replaced with foreground colour. Formulae for this: background * transparency*foreground + (1.0 - transparency)*background Repeated for RGB. Its also worth considering the shape of the surface. For example, a sphere could be more transparent at the centre, becoming less transparent to the edges. Or you could fade transparencies. This will take quite a bit of memory though. ---------------------------------------------------------------------------- Putting It All Together Now you need to put all this together. Your code might look something like: Precalculate as much as possible For every shade For every colour Determine target colour Colour = BestFit(target colour) Store Colour End End Its a good idea to pre-calculate this, and store it to a file, because it can take quite a while to run through. Also, consider the alignment of such tables, and your texture maps. If they are aligned to 64k boundaries, we can improve our routine greatly. We can use the upper 16 bits of the address as the address, and the lower 16 bits as the index: mov ebx, [pointer] mov bl, [u] mov bh, [v] Loading values into BH has the effect of an automatic multiplication by 256. Remember to insert 3 instructions between loading BL/BH, and actually using them - or you will cause an AGI. Tom Hammersley, tomh@globalnet.co.uk [Image]