| DVD Video Reviews |
| Bjoern Roy , (last update November 23th 2003) |
|
The purpose of this site is not to be yet another DVD review site. First of all, i won't comment on the content of the movie. Taste in movies and filmmaking is a completely subjective matter, so why would you care what i like? I try to focus on things that are objectively measurable. That doesn't mean that i won't include a subjective comment from time to time. Its just that i try to backup my analysis with graphs in the audio domain (see my SPL page here) or with screenshots in the video domain (which this page here is dedicated to), to give you an idea what i am talking about. The one important thing i want people to understand is this. I don't take screenshots of an arbitrary DVD, look with a magnifying glass whether i can find someting that is wrong and proclaim its inferiority afterwards. Its actually the other way around. If i watch a DVD on my very revealing setup (see here), my critical eye sees any imperfections within seconds of looking at it. This includes: detail horizontal/vertical, edge enhancement hor/vert, shadow detail, shadow delineation, contrast, black level, color tone/saturation etc... So basically, my personal analysis is finished almost instantly. The only reason why i go through all the trouble of making the screenshots and presenting them in a comprehensible way, is to backup my review of that DVD's picture quality with easy to understand evidence of what is readily apparent on my system. This won't be the usual review page where lots of new DVD releases get reviewed. I will focus on 'controversial titles', like:
Whenever possible and appropriate, i will try to post A/B type screenshots. E.g. on a re-release, i will try to do old/new comparissons. |
| Titles in the Spotlight |
|
| Screenshot 1-0-1 |
| Showing a downsampled or highly recompressed alteration of a screenshot
would obviously completely miss the point. But even showing the raw
screenshots would be a compomise. Why?
Well, DVD frames consist of only 720x480 pixels for NTSC. People often argue that showing these frames upscaled to any resolution above that (1024x576, 1280x720 or 1920x1080) is a waste of time, because "there isn't more information than 720x480 anyway" or "you can't create detail from nothing". While both statements are true, they COMPLETELY miss the point! Upscaling 720x480 to a higher resolution is not about creating 'detail' that isn't there in the first place. Its about presenting those 720x480 pixels on a display device with the least possible amount of high frequency aliasing noise. Huh? Ok, only because you have 720x480 SAMPLES of an analog medium (like film in this case), doesn't mean that displaying them AS SQUARE PIXELS WITH SHARP BOUNDARIES on a display device is the right form of output to reapproximate the smooth analog waveforms it originally captured! By upscaling digital samples to a higher resolution, you interpolate a smooth waveform between the discrete samples. The more interpolation steps the better. The ideal would be to use a sinc filter that yields a sine wave, but thats currently too demanding processing power wise. A typical interpolation that is used is bicubic filtering, which is already better than bilinear filtering, but not quite spline or sinc quality. If you are a bit familiar with digital information processing in the audio domain, you will see that this is exactly the same concept that is used in DACs in CD players for years. Its called oversampling. Think of the same thing in 2 dimensions and you are there. So, again. Displaying 720x480 samples as square pixels means that the difference between the square waveform and the ideal sine waveform is noise that masks the actual detail in picture. This is what we perceive as 'aliasing'. The harsh edge of the square yielding very high noise in the frequency domain (fourier). That is why watching a DVD on a high resolution device (e.g. 9" CRT) at a high resolution like 1920x1080 does actually reveal MORE detail (of that original 720x480) than a 800x600 digital device, because actual detail is masked by less aliasing noise that distracts the eye. Thats why i use the following strategy when doing screenshots analysis. I scaled the 720x480 grabs to 1920x1080 also with a bicubic filter to mimic what a Faroudja 3000/5000 or HTPC would do. To make my own comparissons, i watch these images on my monitor (which is capable of resolving 1920x1080) with a program that is able to switch between 2 images almost instantly (<1/30 sec, AcdSee). To present the differences to others, like here on the web, i crop out certain slices to highlight a certain issue (EE, noise, clipping, saturation). As easy as that, no magic.
|
Back to Main Page