Thesis suggestion in facial recognition system

[redfrost60]

GOOD DAY MGA KA SYMBIANIZE

tanong ko lang po, any suggestion po sa pagdedefend ng system na Facial Recognition System,
yung System po kasi namin,, dapat may recognition mula doon sa facial system tapos igragrant access dun sa Department or Room,

ano ba kadalasan tinatanong sa mga may Facial Recognition System,,, may nagsabi kasi samin dati na tinatanong daw yung accuracy nung system
kaya medyo mahirap daw to, pero yung sa mga ibang school naman sa ibang bansa, lagi nilang ginagamit.. So yung lang tanong ko, any suggestion. thanks

 

[calbee]

Accuracy of your face recognition highly depends on the algorithm and the amount of training input data used.

If you will be using OpenCV there are currently 3 available algorithms you can experiment with:

1. Eigenfaces
To achieve high accuracy in recognizing features (a face in this case) you will need at least 10 training images of that person. The images must be very clear and must have sufficient lighting. Of course, more images mean much higher accuracy.

2. Fisherfaces
This is a discriminatory analysis algorithm that finds facial features that are unique to each person. Just like Eigenfaces, it requires a good amount of training images. The images however are okay to have some missing components such as lighting. In short, this algorithm can work even at night time.

3. Local Binary Patterns Histograms
This algorithm is more complex than the two but more robust because 1) just like Fisherfaces it doesn't need good lighting 2) Image can be scaled 3) Image can be rotated. The penalty is it takes more calculations to find the features by going through each pixel and it's 8 neighboring cells, then do a threshold to return a single value that are then interpolated to retrieve their spatial information.


All of the above OpenCV algorithms can predict the accuracy of each image by passing double type to the third parameter of the predict() function.

Trivia: The first face recognition system used in London subways was easily fooled. People learned that if you tilt your head 45 degree sideways, the system will not be able to detect their faces. This was more of a face detection problem though not recognition.

Also, if you find your recognizer is very slow you should consider buffering the matrices for one second or two (30-60 matrices/fps). There will be a delay in viewing the real-time feed but it will not be choppy.

 

[redfrost60]

Accuracy of your face recognition highly depends on the algorithm and the amount of training input data used.

If you will be using OpenCV there are currently 3 available algorithms you can experiment with:

1. Eigenfaces
To achieve high accuracy in recognizing features (a face in this case) you will need at least 10 training images of that person. The images must be very clear and must have sufficient lighting. Of course, more images mean much higher accuracy.

2. Fisherfaces
This is a discriminatory analysis algorithm that finds facial features that are unique to each person. Just like Eigenfaces, it requires a good amount of training images. The images however are okay to have some missing components such as lighting. In short, this algorithm can work even at night time.

3. Local Binary Patterns Histograms
This algorithm is more complex than the two but more robust because 1) just like Fisherfaces it doesn't need good lighting 2) Image can be scaled 3) Image can be rotated. The penalty is it takes more calculations to find the features by going through each pixel and it's 8 neighboring cells, then do a threshold to return a single value that are then interpolated to retrieve their spatial information.


All of the above OpenCV algorithms can predict the accuracy of each image by passing double type to the third parameter of the predict() function.

Trivia: The first face recognition system used in London subways was easily fooled. People learned that if you tilt your head 45 degree sideways, the system will not be able to detect their faces. This was more of a face detection problem though not recognition.

Also, if you find your recognizer is very slow you should consider buffering the matrices for one second or two (30-60 matrices/fps). There will be a delay in viewing the real-time feed but it will not be choppy.

I see. May i ask lang sir, pano kapag tinanong po kami ng panel .. kasi kapag natapat po ng picture na galing sa cellphone o kaya picture na may mukha nung tao narerecognize po. pano kaya po yun sir ? ano po maiaadvise niyo po ?

 

[calbee]

kasi kapag natapat po ng picture na galing sa cellphone o kaya picture na may mukha nung tao narerecognize po

You mean nade-detect, hindi nare-recognize. Remember, detection is different from recognition in our context.

Ibig sabihin nun gumagana yung face detection algorithm. Kahit mag draw ka ng mukha ng tao sa papel at ilagay mo sa harap ng camera madedetect pa rin yan.

Sa OpenCV yung primary algorithm na ginagamit sa pag detect is Haar Cascades classifier.

Basic idea is you have a separate trainer program that you feed with positive and negative images. Positive: images na may mukha ng tao. Negative: images na walang mukha ng tao. Number of positive images must be more than negative images. Preferably 2x more. The result is then saved to a file.

Then yung classifier you will feed it input from a different source-- a video camera for instance, it will scan each frame beginning upper-left corner of the image with 20x20 bounded square pixels at a time until it reaches the lower-right of the image.

Habang nags-scan, hinahanap nya yung features na nagmamatch sa images na na-train. One feature na nags-stand out sa mukha ng tao are the eyes. Yung mata are usually darker in color compared to the bridge of our nose or compared to forehead or cheeks. So yung algorithm kapag ang mata nasa loob ng 20x20 window, it will know that there is a match. It will then try to match that frame with other features in the training set, gaya halimbawa ng shape ng mukha, or lapad ng forehead, lips, nose, etc. Remember, mukha ang target dito so lahat ng features ng mukha ng tao kasama sa positive training set.

Face detection is actually the easy part because you almost always get accurate results basta ba magandang quality ng training sets mo, compared to face recognition where it uses a different algorithm like those I mentioned in my previous post.

Go to OpenCV's website meron silang mga overview ng algorithm. If you want to go deeper with face recognition and meron access ang school mo sa IEEE/IET Electronic Library or Association of Computing Machinery (ACM), you should log on there. There you'll find dozens of peer-reviewed publications/journals related to Computer Vision.

 

[redfrost60]

You mean nade-detect, hindi nare-recognize. Remember, detection is different from recognition in our context.

Ibig sabihin nun gumagana yung face detection algorithm. Kahit mag draw ka ng mukha ng tao sa papel at ilagay mo sa harap ng camera madedetect pa rin yan.

Sa OpenCV yung primary algorithm na ginagamit sa pag detect is Haar Cascades classifier.

Basic idea is you have a separate trainer program that you feed with positive and negative images. Positive: images na may mukha ng tao. Negative: images na walang mukha ng tao. Number of positive images must be more than negative images. Preferably 2x more. The result is then saved to a file.

Then yung classifier you will feed it input from a different source-- a video camera for instance, it will scan each frame beginning upper-left corner of the image with 20x20 bounded square pixels at a time until it reaches the lower-right of the image.

Habang nags-scan, hinahanap nya yung features na nagmamatch sa images na na-train. One feature na nags-stand out sa mukha ng tao are the eyes. Yung mata are usually darker in color compared to the bridge of our nose or compared to forehead or cheeks. So yung algorithm kapag ang mata nasa loob ng 20x20 window, it will know that there is a match. It will then try to match that frame with other features in the training set, gaya halimbawa ng shape ng mukha, or lapad ng forehead, lips, nose, etc. Remember, mukha ang target dito so lahat ng features ng mukha ng tao kasama sa positive training set.

Face detection is actually the easy part because you almost always get accurate results basta ba magandang quality ng training sets mo, compared to face recognition where it uses a different algorithm like those I mentioned in my previous post.

Go to OpenCV's website meron silang mga overview ng algorithm. If you want to go deeper with face recognition and meron access ang school mo sa IEEE/IET Electronic Library or Association of Computing Machinery (ACM), you should log on there. There you'll find dozens of peer-reviewed publications/journals related to Computer Vision.

I know po sir, meron na rin po kami sir na System na may Facial Recognition in fact sir galing nga po sa OpenCV yung mga algorithm namin.
Sorry Sir, Medyo nagkakalituhan po tayo ,,, yung tanong ko po sir is .. Ano po pwede namin ipanglaban sa panel namen if ever po tinanong kami na..
"Bakit nakakarecognize din ng mukha yung system niyo kapag tinapatan ng Picture from Cellphone, from Pictures (yung mga pinadevelop na picture)"

Kasi sir baka gipitin po kami sa ganyang question, di po namin alam ano po puwede ipaglaban sa panel namin.. kasi sir yung system namin more on security.

Thanks po sir.

 

[calbee]

Well, kung naiinintindihan mo yung face detection algorithm na in-explain ko in a nutshell, specifically yung scanning process, then hindi ka na mamomroblema sa tanong na: "Bakit nakakarecognize din ng mukha yung system niyo kapag tinapatan ng Picture from Cellphone, from Pictures (yung mga pinadevelop na picture)".

Haar Cascades classifier will not know and will not discriminate whether it is looking at a photo, photo of a photo, photo on a cellphone, photo on a paper, developed photos OR real people's faces in front of the camera. As long as they are inside the view of the camera they are all treated equal during the scanning of each frame.

All the classifier does is find a face inside each frame.

When I say frame, I mean camera frame from FPS. FPS stand for frames per second, a measurement for how many unique consecutive images a camera can handle each second. Low-end digital still cameras typically have a frame rate of 1fps. Low end digital video cameras typically have a frame rate of 30fps.

There's no other way I can think of how to explain the process in your defense aside from describing how the algorithm works. I suggest you also study the math behind it and memorize the formulas in case you are asked for mathematical proof.

 

[maritesmacute]

i made this thingy in for student.. all i can say is.. THIS IS NOT ACCCURATE.. panelist has a pictures in their wallet.. gues what? system will recognize them at mark their account as PRESENT in attendance.. but they don't fail, because its just a thesis.. even there's a bug thing in Attendance Face Recognition shit, they suggest that so they are the one who's responsible for Bug xD

btw, the most accurate is Finger Print Sensor xD

- - - Updated - - -

btw OpenCV is really easy thing.. if you are just going to study xD

 

[redfrost60]

Well, kung naiinintindihan mo yung face detection algorithm na in-explain ko in a nutshell, specifically yung scanning process, then hindi ka na mamomroblema sa tanong na: "Bakit nakakarecognize din ng mukha yung system niyo kapag tinapatan ng Picture from Cellphone, from Pictures (yung mga pinadevelop na picture)".

Haar Cascades classifier will not know and will not discriminate whether it is looking at a photo, photo of a photo, photo on a cellphone, photo on a paper, developed photos OR real people's faces in front of the camera. As long as they are inside the view of the camera they are all treated equal during the scanning of each frame.

All the classifier does is find a face inside each frame.

When I say frame, I mean camera frame from FPS. FPS stand for frames per second, a measurement for how many unique consecutive images a camera can handle each second. Low-end digital still cameras typically have a frame rate of 1fps. Low end digital video cameras typically have a frame rate of 30fps.

There's no other way I can think of how to explain the process in your defense aside from describing how the algorithm works. I suggest you also study the math behind it and memorize the formulas in case you are asked for mathematical proof.

Well Sir Thank you, Bale idadagdag nalang talaga namin sa limitation ng system yung pagiidentify na yan .

i made this thingy in for student.. all i can say is.. THIS IS NOT ACCCURATE.. panelist has a pictures in their wallet.. gues what? system will recognize them at mark their account as PRESENT in attendance.. but they don't fail, because its just a thesis.. even there's a bug thing in Attendance Face Recognition shit, they suggest that so they are the one who's responsible for Bug xD

btw, the most accurate is Finger Print Sensor xD

- - - Updated - - -

btw OpenCV is really easy thing.. if you are just going to study xD

Haha, Finger Print nga sana sir gusto namin, kaso mahal yung finger print scanner tapos masyado na daw narecycle yung idea na yun sa school namin at sa ibang school. but anyway Thanks sir, kahit papaano pinaluluwag mo loob ko sa "but they don't fail, because it's just a thesis" hahaha.

 

[qazwsxe]

ano po ba pinagkaiba ng face detection at recognition? kaya po ba to kung simple web cam lang gagamitin? haha ganyan din ksi thesis topic nmin, per 50-50 kme haha

 

[goct8]

sir pautolong nman dyan sa inyo sir
sir panu po ba e scan ang image na nasave q
vb2008 mysql
face recognition..

 

[scarletamora8888]

Ask ko lng kung magkano ung medium pang facial recognition, I mean ung pinaka hardware nyan, I think mahal yan at daming butas sa defense.

Also, ask ko lng kung meron na kayong facial recognition device if meron na san kayo bumili?

if wala pa kayong mga hardware ng thesis nyo for that project, much better if mag change nlng Development sa Agriculture bihira yan gnyn topic

 

[rihom]

Yung thesis ko ay "Face Recognition in Class Attendance". Bali C# gamit ko at Eigenfaces or Eigenvalues (OpenCV).
If I remember correctly, kadalasang tanong talaga ay yung accuracy at ilang trainers ang ilalagay at ano yung hardware requirements. Tinanong din ako dati kung magkano yung Cost kung eimplement sa school.

 

[jarence24]

Yung thesis ko ay "Face Recognition in Class Attendance". Bali C# gamit ko at Eigenfaces or Eigenvalues (OpenCV).
If I remember correctly, kadalasang tanong talaga ay yung accuracy at ilang trainers ang ilalagay at ano yung hardware requirements. Tinanong din ako dati kung magkano yung Cost kung eimplement sa school.

good day po....ask ko lang po about sa proposal niyo na "Face Recognition in Class Attendance". nag raspberry pi at raspberry camera v2 po ba gamit nio??medyo nahirapan po kasi kami sa pag tawag ng camera module sa ginawa naming gui... tnx in advance po