Matlab image processing for facial expression

  • Posted
  • Proposals 2
  • Remote
  • #33069
  • Archived
Andrew N.Javier O. have already sent a proposal.
  • 0

Description

Experience Level: Intermediate
ONLY READ THIS IF YOU CAN DO THIS PROJECT ASAP (IN 10 DAYS)

I need a facial expression recognition program written in Matlab language...
Following are the details:

Requirements:

It should classify an image into one of the six basic emotion catogories.

The data to be used should consist of still images of faces. The images will be of frontal shots of the face with differing backgrounds. recognition arising due to facial hair and glasses will be ignored as being beyond the scope of this current project. The face will be the largest skin covered part of the body in the test images. The program should work on normally any image(grayscale ro colour).

The program will function in two steps, first to detect the face in the complex background using Principal Component Analysis (PCA). The second step would involve recognizing the appropriate features in the face to recognize the expression in a given picture. The program will have to automatically recognize the expressions in the given set of still images. The user will use a GUI to interact with the program. The program will have access to training and test set of images which will conform to the image definitions listed above.

The proposed program will be based after testing the Eigen Faces and Feature Extraction + Learning Method. The parameter for the decision is the accuracy of the results produced by the method. The more accurate methods shall be incorporated into the GUI.

method requirement:
a. FEATURE EXTRACTION + LEARNING
Regions of expressions are extracted out first and normalized across the training dataset. For example, the pupils in the detected faces are recognized and all they are normalized with respect to their orientation and distance between each pair. Fixed region around the detected eyes can be cut out and can used for feature extraction. Gabor filters or Haar filters can be used for real feature extraction and a feature selection is performed using an Adaboost algorithm or forward selection algorithm. These selected features are then used as training data in a SVM classifier. The 44 AU\'s available suggest a multi class SVM problem which can be solved by a \'one against all\' strategy. Another multi class classification method called the ECOC can also be tried to see if it gives better results for the given problem.

Document requirement:

Implementing Full code with complete comments., test samples.
A report on the comparison of both the techniques, also giving their detail
strengths and drawbacks, accuracy results, conclusion what\'s the future solution. (all about 2000 words)
Literature review of the currents techniques and their methodology working (about 2000 words)

CODE SAMPLE : http://www.mathworks.com/matlabcentral/fileexchange/23866-eigenfaces-for-expression-detection

NOTE: I NEED AS MUCH COMMENTS IN THE CODE AS POSSIBLE AND NEED EXPLANATION OF THE CODE


Note: it should be on windows platform...
the code sample is just an idea of the output required but it does not mean the same method required.
we need better results.!!

Clarification Board

    There are no clarification messages.