WHiCH MPEG-4 CODEC YiELDS THE HiGHEST FRAME QUALiTY.

First thing I'd like to mention, I did not want to do an encoding comparison between codecs using source material such as a DVD-RIP. I never rip DVD's and therefore such a test would be totally pointless for me as the encoding results would be worthless.

All footage has been captured using a Pinnacle PCTV Rave PCI card (Bt8x8 based). Input source was Analog Free-To-Air Broacast, received using standard rooftop aerial with RG-59 coaxial cable to the Pinnacle PCTV Rave's Tuner. (you can see the broadcasters logo bottom right of the frame.) There have been NO filters ran on the footage, other than Field Swap (and bi-cubic resize for the 384 x 288 tests).

Secondly, if you're at this page via a search engine or an off-site link, then you've missed out on seeing the encoding results. Please read TEST7 results from THIS PAGE To help put the following frame quality comparisons into perspective.


768 x 576 FRAME QUALiTY

This slideshow contains a single frame from each encode at Quality Based 4 and Constant Bitrate 2000kbps. The index on the left shows you which image is currently being displayed (if the javascript slideshow doesn't work for you, you can download or view each image separately from the links at the bottom of this page)

Ok, first up, I chose this particular frame as it contained quite a bit of detail, had some fast moving areas (such as the rotor blades and fire), had some static areas (such as the dirt and rocks) and also contained some slow moving areas (the chopper). This I thought would equal a nice all-round frame.

*notes* Also All images are stored as JPEGS because PNG or BMP are just too large. However comparing these JPEG's to the original BMP's extracted from the AVI's, I can hardly notice any difference with my eyes.


Slideshow viewing of the original frame against each MPEG-4 codec tested

Due to the load times required to pre-load both slideshows, I've decided to make them into separate pages. I hope this helps people with slower connections. Total size of this slide show is 700 + KB.
( This will open a new browser window )


768 x 576 DiSCUSSiON

CBR vs QB4
In my opinion there's no comparison. QB4 from any codec beats CBR 2000 kbps hands down for frame quality.
BEST QB4
Looking at this frame, I think they've all done a pretty good job, and I'm happy with all of them (except X-Vid 26.12.2003 looks pretty soft, but I guess that's the price you pay for producing the smallest filesize). If I take a small section of the frame (the shadow from the explosion for example), Div-x and FFVFW retained more detail. X-Vid seemed to soften over fine detail.

Now if you remember back to the encoding results, Div-X produced the largest filesize, and now we can see why. Using this frame as an example. I'd select Div-X, however if filesize is a factor (and it must be, considering the variation in filesize between the codecs), then Div-X is not the best option. FFVFW would be my NEXT CHOICE. X-Vid produced a slightly smaller filesize than FFVFW or 3-ivx, however it seems to have softened some fine detail in the frame, 3-ivx looks a little sharper. So if being slightly softer doesn't bother you then X-Vid would also be a good selection.

I do prefer the previous version of X-Vid for frame quality, but then again the filesize savings with the new X-Vid means you could increase the quality settings (quantizer value) so that it could equal in quality the other codecs.
BEST CBR
Well, i can't say I'd be happy with any of these when comparing to the QB4 frames, so the only thing I could say would be a higher bitrate would be needed, which would then bring it in line with the QB4 filesizes and frame quality

However if I had to select, then I would say that the Div-X version looks best. Looking at the chopper door. 3-ivx looks the WORST (and has the largest filesize), FFVFW is a little better (and has the second smallest filesize), leaving X-Vid as having the smallest filesize, but the frame quality is slightly inferior to FFVFW (check out the "United States Army" letters on the choppers tail, X-Vid has smoothed them out more than ffvfw. Div-X shows the least number of compression block artefacts.

Comparing against MPEG, All codecs are better, even 3-ivx. MPEG did a good job on the chopper, but the background detail is all missing, for example the men on the hill (top left of frame), all fine detail is completely missing.

Considering that Div-X's filesize is very similar to the other codecs, but it's processing time was shorter, it's definitely the codec to choose if Constant Bitrate Encoding is your preference (using 2000kbps and 768 x 576). It is a different story for 384 x 288.


384 x 288 FRAME QUALiTY

Using the same frame number as the above slideshow, except this time from the 384 x 288 samples. These samples have been bicubic resized to 768 x 576 to make a more direct comparison easier with the original frame and the above frames. (This includes the original frame. To make the comparison fair, this frame was bicubic resized down to 384 x 288 and then back to 768 x 576


Slideshow viewing of the original frame against each MPEG-4 codec tested

Due to the load times required to pre-load both slideshows, I've decided to make them into separate pages. I hope this helps people with slower connections. Total size of this slide show is 700 + KB.
( This will open a new browser window )


384 x 288 DiSCUSSiON

CBR vs QB4
Wow, a shake up. Divx and 3-ivx CBR are both terrible, but FFVFW and X-Vid CBR are remarkably good, and retain far more fine detail. CBR *just* nudges ahead in frame detail than QB4 for quality. I think just like the CBR VS QB4 comparison for the 768 x 576 test, To get QB4 looking as good as CBR for FFVFW or X-Vid, I'd have to lower the Quality Base value to 3.5 or 3 (this raises the quality percentage from 93% to 95% -> 97%), but this would also increase the filesize, but as QB4 produced smaller filesizes than CBR, QB4 I think would still win (for both frame quality and filesize).
BEST QB4
Well, all appear reasonably equal in this test. Each codec demonstrates strengths in some areas of the frame, but shows weakness in other areas. If for example I take the text "United States Army" on the tail of the chopper you can gauge which codec reduces (or softens) fine detail, as the text changes shape or becomes harder to read. Div-x makes the text look more like the original frame. 3-ivx does well just below the chopper (the ground between the skids and the chopper floor). Ffvfw retains more small colour definitions. Take the white section immediately left of the open chopper door, vertically in the middle of this is a dark line, at the top of the dark line is a black square, ffvfw is the only codec that shows this small item correctly. 3-ivx softens it, divx slightly moves it down, and it's completely missing with x-vid. X-Vid 26.12.2003 is considerably lower quality than X-Vid, but it did produce a considerably smaller filesize, so that would allow you to increase the Quality or Quantizer values to increase its quality and still have it compete for filesize aswell.
BEST CBR
Oh boy what a change from the 768 x 576 test. There are massive differences between the codecs. 3-ivx looks the worst, followed pretty closely by Div-X. Div-x retains more detail in the fire than 3-ivx, but both codecs are WAY behind the quality of FFVFW and X-Vid. Both FFVFW and X-Vid did a fantastic job and the frame is almost identical to the original. Check out the floor inside the chopper where it is slightly darker. X-Vid has removed the gradients from the shadows but FFVFW has tried to leave it intact.

Comparing X-Vid to X-Vid 26.12.2003, the new version seems slightly inferior to the older version (but it did produce a smaller filesize). Check out the left edge of the chopper door, you will see with the 26.12.2003 version the door is the wrong shape. The door handle is smaller.

FFVFW is the winner here, but only by a small margin over X-Vid. FFVFW also produced a smaller filesize and was quicker to encode, than X-Vid, but X-Vid 26.12.2003 did beat it for filesize but looks slightly worse than the older X-vid and took longer to encode.


Summary of frame quality between the codecs for each test. This DOES NOT compare the frame quality between tests, The Stars indicate the quality of the frames for that particular test only, SO a 5 star codec for the 768 x 576 CBR 2000kbps does not indicate it's better or worse than a 3 star codec for the 384 x 288 CBR 2000kbps. This only compares each frame from each codec for that particular test only. This does not take into account encoding time or filesize, That is covered on another page.

SUMMARY OF QUALiTY PERFORMANCE
Div-X FFVFW
030927
3-ivx X-Vid X-Vid
26.12.2003
768 x 576 ENCODiNG
QB ENCODiNG
CBR ENCODiNG
384 x 288 ENCODiNG
QB4 ENCODiNG
CBR ENCODiNG

LEGEND
BEST QUALiTY FOR THAT TEST
WORST QUALiTY FOR THAT TEST



QUANTiZATiON TYPE MPEG vs H.263

Using the same frame number as the above slideshow, except this time comparing Mpeg versus h.263 encodes from both 768 x 576 and 384 x 288 samples, for FFVFW and X-Vid codecs only at this stage.


Slideshow viewing of the original frame against each MPEG-4 codec tested

This is not a slideshow, instead it is interactive, so you can select which pictures you wish to load & compare. NO images are pre-loaded, so only those images you select will be loaded.

I hope this helps people with slower connections.
( This will open a new browser window )


MPEG vs H.263 DiSCUSSiON

768 x 576
QB4
Very small increase in frame quality, especially related to fine detail and sharpness. MPEG wins. An additional benefit is the filesizes were also smaller with MPEG.
768 x 576 CBR
BIG differences are seen here. Large increase in fine detail (such as the chopper door); Large increase in sharpness such as the men on the top left hill; and Massive reduction in the compression block artefacts (such as the fire shadow). MPEG wins again. No increase or decrease in filesize, but MPEG was slower to encode. It still wins though.
384 x 288
QB4
Fine increase in sharpness. (such as the men on the top left hill and the fire shadow). Frame is generally better all over. MPEG wins.

X-Vid also shows increase in sharpness, however it also appears to have sharpened the compression artefacts, so it's difficult to say that this frame is better than the h.263 variant. It is better in some areas, but worse in others. For example check out the top of the chopper door and the turbine. Very "noisy" compared to ffvfw (or even the older version of X-Vid). However X-Vid did provide a considerable filesize saving, so that would allow you to increase the quality factor to regain some of the detail loss.
384 x 288 CBR
Very fine increase in sharpness for both codecs. No where near as much variation as seen in the 768 x 576 test, but MPEG still nudges slightly ahead for sharpness.


If anyone would like to see other tests conducted, or have any input on these tests or my conclusions, or if you have spotted a problem or error, please feel free to Contact Me.

If the Javascript slideshow doesn't work for you, or If you would prefer to view each image separately you can from the below links. Or right click and select "save target as"


iMAGES USED iN THE FiRST TWO SLiDESHOWS
768 x 576 Encoding 384 x 288 Encoding
Pic Video Q18
Div-X QB4
FFVFW QB4
3-ivx QB4
X-Vid QB4
Div-X CBR
FFVFW CBR
3-ivx CBR
X-Vid CBR
X-Vid CBR (031226)
Pic Video Q18
Div-X QB4
FFVFW QB4
3-ivx QB4
X-Vid QB4
Div-X CBR
FFVFW CBR
3-ivx CBR
X-Vid CBR
X-Vid CBR (031226)
iMAGES NOT USED iN THE ABOVE TWO SLiDESHOWS
X-Vid QB4 (031226)
X-Vid QB4 (031226) TURBO
MPEG-1 CBR 2000kbps
X-Vid QB4 (031226)
ADDiTiONAL iMAGES USED iN QUANTiZATiON COMPARiSONS
FFVFW QB4
X-Vid QB4
FFVFW CBR
X-Vid CBR
FFVFW QB4
X-Vid QB4
FFVFW CBR
X-Vid CBR


This Web Site, requires
Microsoft Internet Explorer with JavaScript And Frames Support
Best viewed in 1024 x 768 resolution

This Page was
Created on: 10th January 2oo4
Last Updated on: 19th September 2oo4

(C) Narler
If you didn't get to this page from my Main Page, Click here