Hi Sobakava,
a 8x8x8 RGB reconstruction from a 1280x1024 image creates (8+8+8)x1280x1024 bit array...
= 1280x1024 pixels of 24 bits
= 3.932.160 bytes :wink:
(this site explains some basics : **broken link removed** )
Indeed, the neighbours averaging is the simplest method. It give the worst results... Choose an algorithm in relation to your quality needs.
Most of algorithms may be implemented in Real-time with line FIFO or circular buffer. However, I've never had to do RT code.
According to your target application, you will need contrast correction or other calculation that needs the whole image... Maybe, you could implement complex color reconstruction and others calculatino on the driver level of your target platform (PC?...). A benefit of this solution is the amount of data to transfer, 3 times smaller !