请大家帮忙分析一下这张图的缺陷是怎么引起的

chengsean

To chiefhe:
I know you must have encoutered this problem before.
I am curious about that the sensor can really output YUYV more than 640 pixels ?
If I set the output size to 640 x 480 YUYV,  it output 642 x 480 YUYV.
Is this right ?   

I am not challenging you. I just very curious this can work well ?
Because I am a ISP designer, this is not permited in my design ?

chiefhe

<DIV class=quote><B>以下是引用<I>chengsean</I>在2007-6-21 17:25:23的发言：</B>
To chiefhe:
I know you must have encoutered this problem before.
I am curious about that the sensor can really output YUYV more than 640 pixels ?
If I set the output size to 640 x 480 YUYV,  it output 642 x 480 YUYV.
Is this right ?   

I am not challenging you. I just very curious this can work well ?
Because I am a ISP designer, this is not permited in my design ?

</DIV>


至少高通平台就可以，而且高通会要求多数出几个行同步。
其它我不敢说，但只要自己带有计数器的后端对这些都无所谓。
我调试的sensor几乎都不是正好640x480，多几个没关系，后端不采样就行了。

chengsean

<DIV class=quote><B>以下是引用<I>chiefhe</I>在2007-6-21 19:34:48的发言：</B>



至少高通平台就可以，而且高通会要求多数出几个行同步。
其它我不敢说，但只要自己带有计数器的后端对这些都无所谓。
我调试的sensor几乎都不是正好640x480，多几个没关系，后端不采样就行了。</DIV>


Are you really sure the sensor ouput YUV format ?
Which sensors did you adjust ? Could you give those sensor type ?

William

第一次听说锯尺和PCLK 对YUV采样有关系,    学习学习     ,chiefhe 有联系方式吗  交流交流

chengsean

I do some experiment about pixel delay of YUYV data.
Here are my expierment result.
http://homepage.seed.net.tw/web/chengsean

In the first image, the input is YUYV data and no pixel delay.
It is Y0 U0 Y1 V0 , Y2 U1 Y3 V1.
And the second image the input is also YUYV data and I shift one data to simulate the
pixel delay.
So the data will be U0 Y1 V0, Y2, U1, Y3, V1 .
I ignore the the last line, so you can see a strange line on each end of line.

I had known what chiefhe say. I think the effect chiefhe describe is caused by ISP.  
Thanks very much, I have learned very much.

longago

学到不少,顶一下,让更多的人知道[em08]
我眼神好像不好,楼上给的两副图我就没看出区别~
[em10][em10]

chengsean

<DIV class=quote><B>以下是引用<I>longago</I>在2007-6-25 10:14:08的发言：</B>
学到不少,顶一下,让更多的人知道[em08]
我眼神好像不好,楼上给的两副图我就没看出区别~
[em10][em10]</DIV>


Actually.. Those two images are only a little difference...
What I mention is
If the original data is Y0 U0 Y1 V0, Y2 U1 Y3 V1, Y4 U2 Y5 V2 ..
If we only delay one pixel for YUYV data, the output data will be U0 Y1 V0 Y2, U1 Y3 V1 Y4.
And we know the output data is not YUYV format, it is UYVY data format.
So we convert the format to correct format to YUYV, we can get Y1 U0 Y2 V0, Y3, U1 Y4 V1...
Consider the Y component. If the original first line Y data is Y0 ~ Yn. Then the delay
one pixel Y data will be Y1 ~ Yn+1. You can find that the sequence of the Y is still
the same. Hence the whole image should be looked the same as orignal image but
one pixel shift. But U and V data will have some error. Originally the U0 is for y0, y1.
After pixel delay, U0 is for Y1 and Y2. So there are only a little difference in the image.
The reason cause the zigzag that chiefhe say should be caused by the sensor ISP
internally. Sensor ISP should get wrong bayer pattern for itself.

flie

虽然还是不太明白，但是忍让学到很多！[em08]