Hi everyone,<div><br></div><div>During the Summer of KDE last year, I had implemented HMM face recognition as a part of the libface library (guided by Alex and Marcel). But the problem was scaling it by addition of new faces. The model parameters had to be regenerated every time a new face was added and the parameters changed accordingly. This is actually not very feasible method.</div>
<div><br></div><div>I have done an extensive study on face detection/recognition algorithms and also worked on improving the existing face detection algorithms in libface to make them scale, illumination and pose invariant. I am currently not in college anymore so cannot apply for GSoC this year. But I would like to work on the face recognition part with someone as this is a feature that has the most demand. Any other ideas/algorithms for a scalable and efficient face recognition algorithm can also be looked into in detail.</div>
<div><br></div><div>Thanks,</div><div>Amey<br><br><div class="gmail_quote">On Mon, Apr 2, 2012 at 4:06 PM, Ananta Palani <span dir="ltr"><<a href="mailto:anantapalani@gmail.com">anantapalani@gmail.com</a>></span> wrote:<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">Hi Aniket,<br><br>I think you could consider it, but be sure to read papers by other authors to see if the method is actually as good as the authors describe. Original authors of a method often show their method to be better than the competition (otherwise they wouldn't publish!), but if another author extends a given method, or does a review of all available methods, you can often discover whether the original claims are justified (you can usually find reviews / extensions by looking at articles that cite the original paper).<br>
<br>From what I remember laplacianfaces is a good algorithm, but is sometimes inconsistent in its quality. There is also a 2d-laplacianfaces that performs even better, I believe. I can find the paper if you are interested.<br>
<br>Also, compare the computation time between fisherfaces and laplacianfaces to help you decide which to implement.<br><br>There's also no reason you couldn't implement both methods, if you had time!<br><br>Best,<br>
-Ananta<div class="HOEnZb"><div class="h5"><br><br><div class="gmail_quote">On Fri, Mar 23, 2012 at 9:58 PM, Aniket Handa <span dir="ltr"><<a href="mailto:atneik@gmail.com" target="_blank">atneik@gmail.com</a>></span> wrote:<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
Hi all!<br>
<br>
Should we consider LaplacianFaces for face recognition, because some<br>
papers[1] suggest it gives less false positives as compared to<br>
Fisherface/LDA and PCA. Or are we fine with FisherFace?<br>
<br>
Regards,<br>
Aniket<br>
<br>
Attached:<br>
[1] Xiaofei He; Shuicheng Yan; Yuxiao Hu; Niyogi, P.;<br>
Hong-Jiang Zhang: "Face recognition using Laplacianfaces" in Pattern<br>
Analysis and Machine Intelligence, IEEE Transactions, March 2005<br>
<div><div><br>
<br>
On Mon, Mar 19, 2012 at 10:42 PM, Ananta Palani <<a href="mailto:anantapalani@gmail.com" target="_blank">anantapalani@gmail.com</a>> wrote:<br>
> I don't remember any off-hand, but I do remember they involved haar wavelets<br>
> and Gabor transforms. Also, I don't think they are very accurate without<br>
> video input, but perhaps once sufficient images are tagged as belonging to a<br>
> given individual, it may be possible for the algorithm to determine the<br>
> expression much easier.<br>
><br>
> -Ananta<br>
><br>
><br>
> On Mon, Mar 19, 2012 at 3:28 PM, Francesco Riosa <<a href="mailto:francesco@pnpitalia.it" target="_blank">francesco@pnpitalia.it</a>><br>
> wrote:<br>
>><br>
>><br>
>> Out of interest, do you know of algorithms able to extract facial<br>
>> expressions #1 ?<br>
>><br>
>> #1 <a href="http://en.wikipedia.org/wiki/Facial_Action_Coding_System" target="_blank">http://en.wikipedia.org/wiki/Facial_Action_Coding_System</a><br>
>><br>
>> Il 17/03/2012 18:12, Ananta Palani ha scritto:<br>
>>><br>
>>> Hi Aniket,<br>
>>><br>
>>> I'm not in charge of this particular area of digiKam, but here are some<br>
>>> links you can start looking into to get familiar with using OpenCV and<br>
>>> performing facial recognition, as well as descriptions and papers for<br>
>>> various algorithms. The current algorithm (eigenfaces) used in digiKam<br>
>>> would not be very good with real-world images (even if it worked<br>
>>> properly). One of the best (if not the best) algorithm available would<br>
>>> be the Fisherfaces or Linear Discriminant Analysis (LDA) method. You can<br>
>>> read a nice summary and analysis here:<br>
>>><br>
>>> <a href="http://www.bytefish.de/blog/fisherfaces" target="_blank">http://www.bytefish.de/blog/fisherfaces</a><br>
>>><br>
>>> Here are some good summaries of various algorithms that are available<br>
>>> for further comparison:<br>
>>><br>
>>> <a href="http://opencv.willowgarage.com/wiki/FaceRecognition" target="_blank">http://opencv.willowgarage.com/wiki/FaceRecognition</a><br>
>>> <a href="http://www.face-rec.org/algorithms/" target="_blank">http://www.face-rec.org/algorithms/</a><br>
>>><br>
>>> I believe you have to apply and be accepted to the GSoC program before<br>
>>> you can actually be work on GSoC projects, but someone else will<br>
>>> probably contact you about that (or info might be on the GSoC website?).<br>
>>><br>
>>> Best,<br>
>>> -Ananta<br>
>>><br>
>>><br>
>>> On Sat, Mar 17, 2012 at 2:48 PM, Aniket Handa <<a href="mailto:atneik@gmail.com" target="_blank">atneik@gmail.com</a><br>
>>> <mailto:<a href="mailto:atneik@gmail.com" target="_blank">atneik@gmail.com</a>>> wrote:<br>
>>><br>
>>> Hello there,<br>
>>><br>
>>> My name is Aniket Handa, currently enrolled in Jaypee Institute of<br>
>>> Information Technology, Noida, INDIA pursing B.Tech-M.Tech Dual<br>
>>> program in Computer Science & Engineering. My interests range from<br>
>>> Computer Vision, Computer Graphics to developing Android Apps. I am a<br>
>>> firm believer of FOSS.<br>
>>><br>
>>> I am particularly interested in "Facial Recognition" partially due to<br>
>>> its high demand (Bug 271679: 492 votes) and partially due to my<br>
>>> interests in vision. I saw that this Idea was also in GSoC 2010, so<br>
>>> that puts up questions like: was it taken by some student? If so, were<br>
>>> there some problems?<br>
>>><br>
>>> I am currently looking more about the project and will get back to you<br>
>>> soon. Looking forward for being a valuable part of the community.<br>
>>><br>
>>> Regards,<br>
>>> Aniket<br>
><br>
><br>
><br>
</div></div><div><div>> _______________________________________________<br>
> Digikam-devel mailing list<br>
> <a href="mailto:Digikam-devel@kde.org" target="_blank">Digikam-devel@kde.org</a><br>
> <a href="https://mail.kde.org/mailman/listinfo/digikam-devel" target="_blank">https://mail.kde.org/mailman/listinfo/digikam-devel</a><br>
><br>
<br>
<br>
<br>
</div></div><span><font color="#888888">--<br>
Aniket Handa<br>
Site: <a href="http://atneik.com" target="_blank">http://atneik.com</a><br>
</font></span><br>_______________________________________________<br>
Digikam-devel mailing list<br>
<a href="mailto:Digikam-devel@kde.org" target="_blank">Digikam-devel@kde.org</a><br>
<a href="https://mail.kde.org/mailman/listinfo/digikam-devel" target="_blank">https://mail.kde.org/mailman/listinfo/digikam-devel</a><br>
<br></blockquote></div><br>
</div></div><br>_______________________________________________<br>
Digikam-devel mailing list<br>
<a href="mailto:Digikam-devel@kde.org">Digikam-devel@kde.org</a><br>
<a href="https://mail.kde.org/mailman/listinfo/digikam-devel" target="_blank">https://mail.kde.org/mailman/listinfo/digikam-devel</a><br>
<br></blockquote></div><br></div>