OpenCV for DSP/GPU, MSDN equivalent for CCS, and more

Posted by Shehrzad Qureshi on Feb 17 2010 under TMS320C6x | TMS320C54x | GPGPU | Imaging   

1. A great business opportunity might await anyone brave enough to attempt a port of the OpenCV (Open Computer Vision) library to a DSP platform.  OpenCV is the lingua franca of "industrial grade" image processing, and it's my sense that the DSP imaging libraries out there pale in comparison to the breadth of features offered by OpenCV.  When writing my book for example, I fielded multiple inquiries about making an OpenCV port a centerpiece of the text.  Two problems arise, however.  First and foremost, the scale of this task is fairly large, and is probably why it hasn't been attempted.  In fact there was some measure of effort undertaken to port certain aspects of OpenCV to CUDA, but those efforts appear to have petered out.  Certain image processing "primitives" are just plain difficult to parallelize properly (e.g. connected components).  Secondly, OpenCV is open-source protected by a BSD license, with certain dependencies containing (L)GPL and various other licenses.  Given the fact that coding and testing a reasonably optimized OpenCV port on to a suitable DSP platform would be a non-trivial task, where would the revenue come from -- assuming that the source would have to be released to the public?
2. In early 1990s and prior to that, Windows IDEs were not dominated by a single entity as is now the case (Microsoft Visual Studio rules the roost here, of course).  Back then, a developer could easily choose a compiler/IDE from Borland and not feel left out in the cold.  I think it's fairly evident that one of the major reasons why Visual Studio became thoroughly dominant was the advent of MSDN (Microsoft Developer's Network) and the help system tied into Visual Studio.  This got me thinking -- why is the help in Code Composer Studio so LAME?  Having done significant amounts of programming in both environments, it never ceases to amaze me just how poor the search is in CCS and how I can never find what I'm looking for.  Now even though TI does not have to worry about any competition in the IDE space, think of how much better things would be if the help in CCS was even half as good as the help in Visual Studio.  I haven't done much with the new CCS v4 so I am not aware if this issue has been addressed at all.
3. I just got in my inbox an invitation for a presentation on GPU Computing with ATI Stream Technology on March 9th.  Since I don't know much about Stream aside from what little I've read (and having somewhat disparaged the technology in a previous post) I'm looking forward to this talk and will report back.
4. Without further ado, I have decided on a CUDA project to walk you through the details on this blog!  What I plan on implementing is frequency-domain filtering of images.  More details to follow, but essentially we will multiply the FFT of an image by some mask, then perform the inverse FFT to filter the image.

Rate this article:
posted by Shehrzad Qureshi
Shehrzad Qureshi is an experienced DSP consultant who has worked in the defense, medical imaging, life sciences, & telecommunications industries. He is the author of Embedded Image Processing on the TMS320C6000 DSP (Springer, 2005) and the inventor or co-inventor on multiple patents spanning a gamut of technical fields. More information can be found at www.squreshi.com.

Previous post by Shehrzad Qureshi: Random GPGPU Musings
all articles by Shehrzad Qureshi

Would you like to be notified by email when Shehrzad Qureshi publishes a new blog?

  

I am fairly new to the DSP world, but come from many years experience working with PC software. The comparisons Shehrzad makes between VS and CCS are spot-on and I am constantly frustrated by how difficult it is to find the information I need.
The MSDN makes it very easy to locate things but with the TI documentation it seems I am constantly forced to read 200 page pdf files. At least it is not just me.

hi
i am new in opencv and ccs and i like to make a program with opencv in ccs but i recieve several error about ccs doesnt know c++ codes

Yepp, It would require much work to port OpenCV to ANSI C, eliminate virtual memory uses and simplify dma involving.

Hello,

My project is to implement an OPENCV algorithm to the TMS320DM6437 board using CCStudio.
my problem is :

How to configure ccstudio with OPENCV?
How to impliment an OPENCV program in DM6437 board ?

Feel free to contact me if you have an idea, I'm really stuck!!!

Thank you in advance.

Hai,
I am measuring surface of lateral ventricle from mri image .Please send me some ideas regarding the this problem with code .Now I switched to contour extraction and centroid detection calculation of distance b/n centroid to contour . please help me in this ..

Hi,
great article.
I would like to emphasis that I'm not familiar with OpenCV (yet) - I'm just learning about the field of GPUs.
But, I know (little) about licensing and regarding OpenCV BSD nature (with related (L)GPL and such) I'd like to say that it is not mandatory to release source code under BSD license - LGPL also does not demand to release a sources while GPL is more demanding in this respect (I think original BSD is not compatible with GPL anyways) - it's the subject that would be worth of further investigation for sure (at least industry is for sure interested in this question).

Hi,
I’m new to openCV.
I wanted to know if openCV algorithm could be implemented using Code Composer Studio v3.3.
I’ve already run my program using VC++ 6.0 its working fine.
But when i run the same program using CCS v3.3 it gives me following error during building process…

Linking..
>>error: library ‘C:\\Program Files\\OPenCV\\lib\\cxcore.lib’, member ‘cxcore100.dll’ is of unknown type

Has anyone compiled openCV using CCS and got same error?
Please let me know how to solve this problem

Thanks