• State
  • Locked Locked
  • Replies 7 replies
  • Subscribers 100 subscribers
  • Views 324 views
  • Users 0 members are here
Support feedback
Options
Options
Similar topics

This thread has been locked.

If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.

Original question:

TDA4VM: purple edge problem on TDA4

TDA4VM: New findings on the purple edge problem on TDA4

wang zhili
Intellectual 875 points
Part Number: TDA4VM

Hi Gang,

I have some new findings on the purple edge problem on TDA4, and there are new questions in the meantime.

We found that the purple edge problem in RAW image is not so obvious as in YUV image around the hightlight part in image, as shown in the snapshots. The edge of the highlight in red rectangle in RAW image (fig1-1 & fig1-2) have color problem maybe cause by the lens, but the same place in YUV image (fig2-1 & fig2-2) become more serious and some of the pixels even turned black, which makes the purple edge problem looks more obvious. So, do you have any idea on which process in TDA4 will cause this problem?

We tried to turn off EE plugin, but there is no change on the purple edge.

Is this related to the PWL setting?

fig1-1 raw image  fig1-2 partial magnification of fig1-1 (red rectangle)

fig2-1 yuv image

fig2-2 partial magnification of fig2-1 (red rectangle)

The RAW & YUV images are attached here: purple edge problem.zip

Thanks.

  • 0
    TI__Guru 52785 points

    Hi Zhili,

    Are the raw images with knee points below and pedestal 64?

    xp = [0,1023,1279,1535, 2303, 2559, 2943, 3071, 3327, 3455, 3519, 3647, 3775, 3903, 3999, 4095, 65535]
    yp = [0,1023,2047,4095,16383,24575,49151,65535,131071,196607,262143,524287,1048575,2097151,4194303,16777215,16777215]

    In general, YUV is more colorful than RAW since WB and CCM are applied in YUV.
    The problem in your test scene is more complex than simple CA because of the high contrast from specular reflection and WDR tone mapping.
    Tone mapping may make the color worse.

    If you shift the red rectangle to the center of the view, do you see much less color artifacts?

  • 0 in reply to Gang Hua
    Intellectual 875 points

    Hi Gang,

    Gang Hua said:
    Are the raw images with knee points below and pedestal 64?

    Yes, the pedestal has been set to 64, but I changed the knee points to a new one below, which had no noticeable effect on the purple edge problem around the highlight region.

    https://e2e.ti.com:443/cfs-file/__key/communityserver-components-multipleuploadfilemanager/02d1cc68_2D00_c02a_2D00_479c_2D00_9a53_2D00_8ab109abbd5c-499286-complete/OVX3C_2D00_MAX9295_5F00_MARS_5F00_kneepoints.txt

    Gang Hua said:
    Tone mapping may make the color worse.

    I turned off the GLBCE, but the highlight region still have purple edge. 

    Gang Hua said:
    If you shift the red rectangle to the center of the view, do you see much less color artifacts?

    It will less obvious than the case in the side or corner of image, but the color artifacts is still can be seen easily.

    Do you have any advices for how to solve or improve the purple edge problem in YUV image?

    Thanks.

  • 0 in reply to wang zhili
    TI__Guru 52785 points
    wang zhili said:

    I turned off the GLBCE, but the highlight region still have purple edge. 

    Yes, there will be as CA is still present.

    wang zhili said:

    It will less obvious than the case in the side or corner of image, but the color artifacts is still can be seen easily.

    Could you share the raw/yuv images for this with the knee points?
    I cannot access the knee points file you shared above.
    From this we may see if there is something we can do to tune CFAI better.

  • 0 in reply to Gang Hua
    Intellectual 875 points

    There is the RAW&YUV images and the knee points:

    RAW&YUV_images_knee_points_purple_edge.zip

    I also attched the images with the old knee points, you can compare them.

  • 0 in reply to wang zhili
    TI__Guru 52785 points

    Hi Zhili,

    There is a clear purple edge when the phone is around image corner and no purple edges when it is close to image center.
    The CA issue of purple edges should be reduced by CAC.

    Overall, the YUV images look very unnatural with heave posterization.
    You may have too strong noise filtering and/or tone mapping.

    The other issue of the darker band around the specular reflection seems to be related to tone mapping control.
    The transition is very sharp around the reflection.
    I will have to take a deeper look.

  • 0 in reply to Gang Hua
    Intellectual 875 points

    Hi Gang,

    Thanks for your support very much.

    Gang Hua said:
    You may have too strong noise filtering and/or tone mapping.

    The strength of noise filtering is around 1, and the glbce is 255.

    Gang Hua said:
    The other issue of the darker band around the specular reflection seems to be related to tone mapping control.

    Do you have any advice on this? I found that reduce the strength of glbce can impove the darker band, but the whole image will become very dark. Is it a good method that adjust gamma on this basis?

  • 0 in reply to wang zhili
    TI__Guru 52785 points
    wang zhili said:

    The strength of noise filtering is around 1, and the glbce is 255.

    Noise filter strength is relative.
    It depends on your threshold values.
    If you see too much posterization with NSF4 on vs NSF4 off, then you shall lower NSF4 strength at least for bright pixels and bright conditions.

    Check if it is due to NSF4 or GLBCE first.