GET Quantization Matrix Manipulations / Sudo Null IT News FREE

This clause will be of interest primarily to those who, for whatever reason, prefer AVI and XVID, sooner than MKV and H264. The doom9 and ixbt forums raised questions about the role of various matrices, besides as the manipulation of their have, but intelligible answer, besides intensifying the theory of separate cosine transmute, no nonpareil gave it. Thither is an article where dandy matrices are given, but their advantages and disadvantages are not explained in that location. Thus I will be glad if I rotate to be a pioneer.

So, the task was as follows: to select the matrix coefficients in such a style Eastern Samoa to remove quadraticity when encoding with a invariable quantizer> 2. The file size should comprise smaller than with the synoptical quantizer with the common matrix H263.


In order not to torment you with a deeper insight into the nuances of the learn, Hera is a photo of the ensuant matrix, and an account below.

Red squares are the most important parts. They are responsible for macroblocks, which are almost entirely made up of one color. Since at that place are most such blocks in the video stream, both file cabinet size and modality perception of quality devolve on their publicity. Thence, the task was reduced to the selection of such coefficients at which the ratio character / size will be maximum. The unexpended coefficients were set happening the principle of "slightly big than the previous", small details of the mental image were neglected. Particularly important is the number 12 in the second straightforward. With quantizer 4, if you increase this coefficient, then there will be squares in the background, and a step-down leads to a fulgurating step-up in file cabinet size.

A little much about coding features.
To make this matrix effectual, some settings must cost observed. Firstly, IT is necessarily a constant quantizer of 2, 3 or 4. Less than 2 makes no sentiency, and Sir Thomas More than 4 - background distortions appear. No fractional quantizers! Photo 2.

Secondly, additional quantizer settings that appear when you flick on the upper "more" button.

Important settings are circled in blood-red. Quantizer ratio - the multiplier factor of the independent quantizer, indicated earlier (pic 2), is required 1. Quantizer offset - growth of the quantizer for non-key frames, best 2. For maniacs, quality can be 0, for greater compression you can increment, but I do non advise.
I as wel highly recommend disabling Adaptive quantization and Treillage quantization (bottom clitoris "more").

For example, I took a small piece from a movie with good quality, resolution 720x304, initial bitrate 1300 (without audio). Coding options and received bitrate:
matrix H263, quantizer 2 = bitrate 953
my intercellular substance, quantizer 2 = bitrate 883
my matrix, quantizer 4 = bitrate 433

With quantizer 2, it is almost unworkable to rule differences in quality, with quantizer 4 with my matrix the differences are visible with frame-by-frame screening, with quantizer 4 with matrix H263 - squares, squares.

Conclusion: the matrix is ​​usable at any resolutions, you can center on quality or a smaller sized. It is especially utilitarian at low resolutions, since quadraticity is minimal, and in a broad way resolutions with a quantizer of 4, information technology is enough to upload movies to the Internet. At modest resolutions, you can slimly reduce the coefficients in the red square and there will exist candy. And forget about two passes.

It might non suit someone that the details in my matrix are too smooth (all coefficients that are> 20), but on that point are two reasons for this. Firstly, it contributes to good compression and small size, and secondly, the sharpness, as a matter of fact, is not unrecoverable evermor. If the video is encoded with high quality, then during viewing, you can enable the sharpen filter Sharpen. I prefer msharpen from the K-Lite Codec Pack. There you can rattling swimmingly adapt the edge and it works more correctly than others. In all but cases, a repand video with a sharpness filter makes a better impression than if you encode sharpness into a video swarm. This is due to the peculiarities of encoding a signal away quantisation, which I will not discuss in detail.

Now for the effectual VirtualDub filters electric sander and temporal smoother, some intrinsic.
The first is smoothed by the Gauss method acting, so that the picture is more cartoonish. If you use a value within 1-5, past you can successfully smooth unfashionable the squareness and noise at the boundaries of objects. Slows down encryption by 10-20%.
The s one does not smooth, but rather stabilizes the picture, makes the current frame dependent along the early one. The value sets the number of previous frames for approximation. Recommended 1 or 2. If you set> 4, then a wave of the hand will bet like a fan. Indispensable for stabilizing twitching background. Slows down encryption by 30-50%.

In the long run, there is only one thing left to say: Le XVID Est Mort? Vive Le XVID!
Code for health.

DOWNLOAD HERE

GET Quantization Matrix Manipulations / Sudo Null IT News FREE

Posted by: salguerounnot1977.blogspot.com

0 Response to "GET Quantization Matrix Manipulations / Sudo Null IT News FREE"

Post a Comment

Iklan Atas Artikel

Iklan Tengah Artikel 1

Iklan Tengah Artikel 2

Iklan Bawah Artikel