• State
  • Locked Locked
  • Replies 7 replies
  • Subscribers 101 subscribers
  • Views 840 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.

Image processing on the 6416

Lars
Prodigy 60 points

I want to work on a face detection based on colors on the C6416 DSP.

I use a kind of a YUV color space to segment the human color, but I get some noise at the background of some pictures.

Example:

 

As you can see the yellow closets are causing some noise with my approach (and the lighting is kind of bad, so my hand is not fully red). I've implemented some kind of Image Opening (without the C64X Image Lib) to fix this but it's rather slow. Any ideas for a fast method to remove this kind of noise?

I also tried a algorithm which checks the last 22 pixel within the segmentation and sets these pixels red if they match the treshold of 10 pixels. This is faster than segmentation and then opening but I need ~2 million cycles for a 320x240 picture, there must be a faster way though. My aim is a 25 fps video stream...

  • 0
    Prodigy 30 points

    hello friend

    can you give brief description of how you wrote code in this program..because i am also trying to do some project in image processing.

    Regards,

    chandu

     

  • 0
    Intellectual 330 points

    Hi

    there seems to be some overflow errors in your yuv conversion. Recheck the formulas and remeber to cchek for overflow in the Y and U and V

    If you post the code I may be able to help more

    Good luck

    Albert

  • 0 in reply to chandra sekhar37126
    Prodigy 60 points

    chandra sekhar said:

    hello friend

    can you give brief description of how you wrote code in this program..because i am also trying to do some project in image processing.

    Regards,

    chandu

     

    Hello Chandra.

    My work was based on http://ieeexplore.ieee.org/xpl/freeabs_all.jsp?arnumber=1581185 where Melanie Po-Leen Ooi is using the Reversible Color Transform (RCT) (see http://en.wikipedia.org/wiki/JPEG_2000 under Color components transformation) to get the YUV or YCrCb color space. You'll only need the Cr and Cb color space to segment the human skin-color. I don't know if I'm allowed to tell you her ranges, maybe it's better that you get a IEEE subscription.

    Greets

    Lars

     

    P.S. There are no overflows in my YUV conversion. It was bad lighting that caused the "sketchy" color segmentation. For the noise reduction I used a unconventional idea of myself: I used a 8x8 subwindow and simply added the segmented pixels in it. If more than 10 pixels are skin-color I set the whole 8x8 subwindow to skin-color. Otherwise its background like it should be. Its a kind of Image Opening/Closing with a quantization included. It worked for me but it's kind of unconventional I guess..

  • 0 in reply to Lars
    Prodigy 30 points

    Thats great Lars.....i am very excited hearing your project...you are really doing a great job...!!.......Are you using DSK C6416 board of TI...? if so how can you program such a good project like face recognision? i still wondered as i am working on kit but unable to find any such applications on it. I could just do some audio applications... thats it...so can you tell me your hardware stuff involved in your face recognision ?..are u using Xilings board too?..or just a DSK C6416?....please tell me the hardwares associated in your face recognision project...

    Regards,

    chandra sekhar.

  • 0 in reply to chandra sekhar37126
    Prodigy 60 points

    I used the daughterboard DSK-EYE from Bitec on the DSK C6416 for the face detection in my diploma thesis. There a were a few problems with the vertical SYNC signal of the DSK-EYE. I had to program some little changes in the FPGA software. But I guess there are some other daughterboard solutions out there?

  • 0
    Prodigy 10 points

    hi

    i m working on my fyp and need to do image processing through DSP kit. i intend to use C6416 DSP processor for this purpose. Can you guide me which camera to use for this purpose and how to interface camera with the DSP kit?

    i desparately need help in this purpose in urgent basis cuz of lack of time.

    regards...

     

  • 0 in reply to Lars
    Prodigy 70 points

    Hey,

    I am using tms3206713 kit.
    I have a problem in reading an image.
    I believe that the image must be included as a .h file in the C program to read it.

    Can u please guide me to converting an image into a .h file ?

    Regards,
    Yash