Revision history [back]

The first thing I would do is to estimate the actual recognition rate, like shown in this answer:

• FaceRecognizer in OpenCV Python FAR/FRR

If the recognition rate turns out to be too low, it's time to preprocess the images. If your images are subject to differences in illumination you could try the approach given in:

• Tan, X., and Triggs, B. "Enhanced local texture feature sets for face recognition under difficult lighting conditions.". IEEE Transactions on Image Processing 19 (2010), 1635–650. (PDF), (C++ Code), (Python Code)

But what is going to give you the greatest increase in recognition rates is the correct alignment of your image data. The Python script I have given in the mentioned tutorial might be sufficient to manually crop the images, but for an automated system you'll need something more clever. An interesting approach (and code!) is given in:

• Gary B. Huang, Vidit Jain, and Erik Learned-Miller. "Unsupervised joint alignment of complex images." International Conference on Computer Vision (ICCV), 2007. (Project page), (PDF Online available), (C++ Source code)

You'll probably need to make some minor changes to make the code work in a recent OpenCV 2.4, but I think it's a feasible task.

Lately another highly interesting approach to head pose estimation was given in:

• X. Zhu, D. Ramanan. "Face Detection, Pose Estimation and Landmark Localization in the Wild" Computer Vision and Pattern Recognition (CVPR) Providence, Rhode Island, June 2012. (Project page), (PDF Online available), (C++ Source code)

But I didn't have time to experiment with it yet, so I don't know if you can use it in a real time environment. There's also a cool blog post by Roy from morethantechnical.com, which is worth reading:

• morethantechnical.com: Head Pose Estimation with OpenCV & OpenGL Revisited (with code)

I guess that makes a good start for a research!

The first thing I would do is to estimate the actual recognition rate, like shown in this answer:

• FaceRecognizer in OpenCV Python FAR/FRR

If the recognition rate turns out to be too low, it's time to preprocess the images. If your images are subject to differences in illumination you could try the approach given in:

• Tan, X., and Triggs, B. "Enhanced local texture feature sets for face recognition under difficult lighting conditions.". IEEE Transactions on Image Processing 19 (2010), 1635–650. (PDF), (C++ Code), (Python Code)

But what is going to give you the greatest increase in recognition rates is the correct alignment of your image data. The Python script I have given in the mentioned tutorial might be sufficient to manually crop the images, but for an automated system you'll need something more clever. An interesting approach (and code!) is given in:

• Gary B. Huang, Vidit Jain, and Erik Learned-Miller. "Unsupervised joint alignment of complex images." International Conference on Computer Vision (ICCV), 2007. (Project page), (PDF Online available), (C++ Source code)

You'll probably need to make some minor changes to make the code work in a recent OpenCV 2.4, but I think it's a feasible task.

Lately another highly interesting approach to head pose estimation was given in:

• X. Zhu, D. Ramanan. "Face Detection, Pose Estimation and Landmark Localization in the Wild" Computer Vision and Pattern Recognition (CVPR) Providence, Rhode Island, June 2012. (Project page), (PDF Online available), (C++ Source code)

But I didn't have time to experiment with it yet, so I don't know if you can use it in a real time environment. There's also a cool blog post by Roy from morethantechnical.com, which is worth reading:

• morethantechnical.com: Head Pose Estimation with OpenCV & OpenGL Revisited (with code)

I guess that makes a good start for a research!