• State TI Thinks Resolved
  • Locked Locked
  • Replies 6 replies
  • Answers 3 answers
  • Subscribers 56 subscribers
  • Views 930 views
  • Users 0 members are here
Support feedback
Options
Options
Related

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.

Compiler: Bicubic interpolation scan conversion on C64x DSPs

Olivier Rukundo
Prodigy 40 points

Tool/software: TI C/C++ Compiler

I implemented bicubic interpolation for scan conversion of 2/3D ultrasound data, but results obtained do not look better than the results obtained using bilinear interpolation. Has anyone encountered this before? Or can anyone confirm that the bicubic interpolation is an unsuitable algorithm for scan conversion operations of ultrasound data?

  • 0
    TI__Guru**** 171044 points

    Hi,

    Please apologize for the delay. This question is very specific to your application and you tagged it with #Matlab

    Therefore, at first glance it does not seem related to not to any of our tools. What SW and TI processor are you using?

    Regards,
    Rafael

  • 0 in reply to desouza
    Prodigy 40 points
    It is related to Ultrasound Scan Conversion on TI’s C64x+ DSPs (see: www.ti.com/.../sprab32.pdf)
  • 0 in reply to Olivier Rukundo
    TI__Guru** 116170 points
    Olivier,

    This is legacy software that is no longer being maintained by our development team. I am trying to find appropriate expert who can comment on this query regarding this medical library.

    Sorry for the delay. I will try to post a response as soon as I hear back from the expert. In the meantime, can you provide details on what your test case looks like and if this is something that. It might also be useful if you can provide the output from the bilinear and bicubic interpolation and the input to the interpolation for us to comment on.

    Regards,
    Rahul
  • 0
    TI__Intellectual 800 points

    I am not an expert in this library, but if the data you have is obtained using high enough sampling rate to preserve the necessary details, you will have a hard time seeing any improvement with using higher order polynomial interpolation. That is, n-linear interpolation would produce satisfactory results.

    Another possibility is that you did not properly implement bicubic interpolation. (although, I would expect much worse performance in case of bugs...)

    Finally, one way to check this would be to obtain data that is sampled at twice as high rate (if possible) and see if it is linear in nature relative to the original (lower) rate. In other words, you may be able to measure the error of bilinear interpolation.

    This would be related to total energy in the upper band of the signal obtained using higher rate. If the total energy in the higher band is negligible, using n-linear interpolation at half rate (lower band) would not produce noticeable errors.

    In case of images, we would be talking about "spatial frequency". In case of time series, we are talking about regular signal frequency.

    Having said all of this, did you try to objectively measure the impact of bilinear vs. bicubic interpolation? As it was implied above, in some cases you may not be able to "see it", but you may be able to "measure it". If you are "looking" at the resulting image to assess quality, you may be able to process both images with some suitable "objective" functions that could capture relevant image features and try to ascertain how bilinear vs. bicubic perform using those objective measures.

    Not sure if any of this was helpful, but it may give you an idea how to get closer to satisfactory explanation of what you have observed.

    Regards,
    Bogdan

  • 0 in reply to Rahul Prabhu
    Prodigy 40 points

    Thanks, Rahul. I would love to share the output images only if you could provide a confidential email to which I can send them (due to our internal rules, I am not allowed to publicly share such images). Also try to get in touch with Xiaohui Li, since he/she is the one mentioned at www.ti.com/.../sprab32.pdf

  • 0 in reply to bogdank
    Prodigy 40 points

    Bogdan,

    Thank you! I understand that I could objectively measure the impact of bilinear vs. bicubic interpolation. However, it is too early to do that, since I am not yet satisfied with the results obtained using the bicubic interpolation. In brief, the bicubic interpolation produced the sectored image, that looks like shifted to the 'right' and 'downwards' thus creating jaggy-like artefacts in some edge-like areas of the sectored image features. That's why I asked the above question - seeking to know if someone has ever implemented bicubic interpolation successfully for ultrasound scan conversion purposes. If you ever did that, please let me know.