You'd have to be more precise. Memory for what? And as for reading pixels, it depends on your equipment. Some cameras are made especially for microcontroller-based robots, like the CMUcam and the AVRcam. Another possibility is to use a laptop/PDA with a simple webcam, and interface with whatever API your platform support...
So, the short answer is, you need to decide on what platform you're using before you even worry about how to read pixel values. But if you're interested in serious vision research, you're probably going to want to look at a linux/PC based platform, either with an onboard PC and webcams/HD cameras, or a mobile, remote controlled "drone", with a wireless camera system and some video acquisition devices, as someone else here is doing.
I'm personally going with my own laptop, running linux, and using two USB 2.0 webcams. I believe that's the simplest option for a robot of a moderate size (mine will be about 4.5' high).