For most video compressors basically they use a pixel block scheme.
64x64 pixels (why you get that horrible blocky effect at low-res or high-rate compression) which basically samples each pixel around it and if it's within a certain range will then say 'well these pixels are close to X Colour' it'll then adverage them all and put the colour dword (of byte if it's paletted colour) plus the number of pixels. Often it's a 24-bit colour value with an 8-bit offset.
If it's a 16-bit compression then then the choice usually is to just drop the colour range, and deliberately use an LDR.
So you'll find those sample of colours end up being like X colour for X pixels.
This can be up to 128 pixels, so your getting a single dword = 128 pixels which normally would be a 24/32-bit colours meaning up to what, 512 byte down to 4 byte .. in the best case scenario.
Uusually it's not done to that sort of extreme and the whole block will get a single pass gzip or something as well. Then you have the whole huffman compression which just plainly confuses me.