Some more offline comments and testing recently have raised a critical issue; the use of Square Pixel Sampling and Non-Square Sample Formats. Before that though, let's talk about validation testing. When we get to this level of detail especially, it is important to validate our processing, tools, and complete process. Correcting Aspect Ratio is certainly no exception. In fact, I think we would all agree that when this level of detail really matters, you must validate. Lot's of great resources out there about that topic.
Square Pixel Sampling
Have you already done some testing? Did you notice that when you resized 480i NTSC based on the luminance sampling rate methodology, your method of just using the pixel matrices was more accurate down to the pixel (e.g. you just resized 720 x 480 down to 640 x 480)? Ah, well, what if the original source was acutally a square pixel to begin with? What if equipment manfacturers didn't understand this issue? Well then, we would not have non-square samples to begin with; we'd have square 1:1 PAR samples discretely created by the sampling device (DCCTV system) or camera (e.g. Digital transmission). If we then take those square samples and store them in a format and/or codec designed and intended for non-square use, under those circumstances we CAN USE THE PIXEL MATRICES to properly correct Aspect Ratio.
I told you, a lot of variables. Let's get this knocked out and put in place as a Best Practice so there isn't so much confusion in the industry. Thanks everyone, for your feedback to-date.