t******q
发帖数: 117 | 1 【 以下文字转载自 EE 讨论区,原文如下 】
发信人: leizj (lilith), 信区: EE
标 题: Re: 还是请教关于DCT用于图像压缩。这么说吧？
发信站: Unknown Space (Fri Aug 22 09:31:23 2003) WWW-POST
let's say:
Y=A*X*B,
so
X=C*Y*D=C*A*X*B*D
therefore:
C*A should be unit matrix, so C=A^(-1) and A=C^(-1)
B*D should be unit matrix.
这只是基本的变换和反变换必须满足的条件。
如果你还想要你的变换有更多的特性，还得满足其他的条件。 | t******q
发帖数: 117 | 2 【 以下文字转载自 EE 讨论区,原文如下 】
发信人: mizhael (mizhael), 信区: EE
标 题: Re: 还是请教关于DCT用于图像压缩。这么说吧？
发信站: Unknown Space (Fri Aug 22 12:31:18 2003) WWW-POST
Yes, good idea, thank you for pointing them out...
In all standards such as JVT, the authors keep Y=A*X*A' instead of Y=A*X*B,
can you give me some hints why they like Y=A*X*A', if you have time?
Also, Y=A*X*B or Y=A*X*A' are for separable transforms, how about Y=A*X where
A is 64x64 and X is 8x8 reshaped to 1x64 column vector, which is for
non-seperable c | t******q
发帖数: 117 | 3 【 以下文字转载自 EE 讨论区,原文如下 】
发信人: con (earn money), 信区: EE
标 题: Re: 还是请教关于DCT用于图像压缩。这么说吧？
发信站: The unknown SPACE (Fri Aug 22 15:01:54 2003), 转信
其实图象压缩有有损压缩
如果研究新方法的话
不应该把条件限制这么死
不用要求perfect recontruction
所以你的unit matrix等类似的条件把
自己的手脚束缚住了
作研究不是这么干的
其实只要要求恢复的图象和
原来的图象的差别小于一定的threshold
就可以了.这样的变换就多了去了
approximate recontruction vs. perfect reconstruction. | t******q
发帖数: 117 | 4 【 以下文字转载自 EE 讨论区,原文如下 】
发信人: TigerJade (小人莫不有才), 信区: EE
标 题: Re: 还是请教关于DCT用于图像压缩。这么说吧？
发信站: Unknown Space (Fri Aug 22 15:50:35 2003) WWW-POST
my 2 cents. Although you can always treat linear transform in matrix
multiplicatoins, you have no way to implement it in the same manner, due to
the high complexity of the matrix multiplication ~ N^3. That why we love DCT
or DFT, because they are on the order of Nlog(N). If I remember correctly, the
complexity of DWT is KN, where K is the length of the fi | t******q
发帖数: 117 | 5 【 以下文字转载自 EE 讨论区,原文如下 】
发信人: TigerJade (小人莫不有才), 信区: EE
标 题: Re: 还是请教关于DCT用于图像压缩。这么说吧？
发信站: Unknown Space (Fri Aug 22 15:53:02 2003) WWW-POST
Well, I guess, when you do the linear transform, you cannot loss any info in
the transform, because at that time, you don't know which info is useless (or
smaller than the threshold as you said). In the transform domain (with a
compact expression), after the linear transform, you can then figure out what
info is useless or not significant.
Z=C*Y*D... | t******q
发帖数: 117 | 6 【 以下文字转载自 EE 讨论区,原文如下 】
发信人: leizj (lilith), 信区: EE
标 题: Re: 还是请教关于DCT用于图像压缩。这么说吧？
发信站: Unknown Space (Fri Aug 22 16:13:05 2003) WWW-POST
我前面讲了，这些变换在计算机实现的时候因为
浮点精度的问题已经变成有损的了。另外在
应用这些变换的时候不要只是考虑压缩。图像
处理不仅仅是压缩。现在的视频软件还得考虑
zooming, panning, 改变图像分辨虑等等
很多应用。 DCT的选用使很多这些操作都可以
很容易的在频域实现。最终用户只考虑速度
和质量。 现在用户已经在抱怨视频编辑软件
制作 VCD 和 DVD 太慢了。制作一个1小时
的VCD要用大概4小时的时间。而且6-8Mbps
的带宽已经不够压缩 720*480 的DVD了。
不错，做研究可以随心所欲。可是你想想，
图像处理研究了几十年最后大家只认 DCT
和小波。 你想做多少年来搞出一个更好的
变换。而且如果你没法证明你提出来的变换
至少在某方面优于现有的，那怎么毕业呀。
Z=C*Y*D.. | t******q
发帖数: 117 | 7 【 以下文字转载自 EE 讨论区,原文如下 】
发信人: TigerJade (小人莫不有才), 信区: EE
标 题: Re: 还是请教关于DCT用于图像压缩。这么说吧？
发信站: Unknown Space (Fri Aug 22 16:18:04 2003) WWW-POST
Well, if you are doing research on designing transform, it looks like you've
already considered to much, which really limits your idea. If you are doint
research work, you only need to do theorectical analysis. The implementation
part can be done by other guy, who is responsible for choosing the best one
from different algorithms. | t******q
发帖数: 117 | 8 【 以下文字转载自 EE 讨论区,原文如下 】
发信人: kinetic (Tie Tie), 信区: EE
标 题: Re: 还是请教关于DCT用于图像压缩。这么说吧？
发信站: Unknown Space (Sat Aug 23 20:22:33 2003) WWW-POST
Simply speaking, signal compression is all about throwing some transforms to
particular signal sources and getting rid of small coefficients. Therefore,
those transforms actually induce some perceptual models. DCT and DWT
successfully capture some important signal characteristics, especially for
audio and images. Since both DCT and DWT have fixed basis, th | t******q
发帖数: 117 | 9
and
successful.
really
are
images,
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
zoom, rotation, and so on, that is why in latest standards, dirctional
predication are adopted.
【在 t******q 的大作中提到】
: 【 以下文字转载自 EE 讨论区,原文如下 】
: 发信人: kinetic (Tie Tie), 信区: EE
: 标 题: Re: 还是请教关于DCT用于图像压缩。这么说吧？
: 发信站: Unknown Space (Sat Aug 23 20:22:33 2003) WWW-POST
: Simply speaking, signal compression is all about throwing some transforms to
: particular signal sources and getting rid of small coefficients. Therefore,
: those transforms actually induce some perceptual models. DCT and DWT
: successfully capture some important signal characteristics, especially for
: audio and images. Since both DCT and DWT have fixed basis, th
|
|
