FAQ 1: Should we submit one report per group or individual student reports?
ANS: Certain parts of the reports require individual responses; some other parts (example, MATLAB code) do not. Our preference is one report per group, making sure that the individual assignments are properly integrated.
FAQ 2: I did not upload my face image to the database. What should I do.
ANS: Pick a random face from the database, make sure that you note that your face is not in the database, and report the results on the random face image.
FAQ 3: When and where should I turn in the report?
ANS: To the HW box on the fifth floor, by 5PM on Thursday, June 3. Electronic/e-mail submissions will not be accepted, will not be acknowledged, and will not be graded.
GROUP ID#:
GROUP MEMBERS: <NAMES>
ALL MEMBERS UPLOADED THEIR FACE IMAGES? IF NOT, IDENTIFY THOSE WHO DID NOT.
[THIS PART CORRESPONDS TO ASSIGNMENT #2; THIS PART IS COMMON TO ALL GROUP MEMBERS.]
1. IMAGES OF THE FIRST 20 EIGENFACES AFTER SORTING THE EIGENVALUES, WITH THE FIRST EIGENFACE CORRESPONDING TO THE LARGEST EIGENVALUE.
2. A PLOT OF THE FIRST 50 EIGENVALUES.
3. VALUE OF K FOR T_LAMBDA = 0.95, AND FOR T_LAMBDA = 0.99.
4. MATLAB LISTING OF ALL OF THE CODE YOU DEVELOPED FOR THIS PART. Include all of the MATLAB listings in one place, at the end of the report. Clearly identify the individual sections. For example, number the MATLAB code for this part as: Section 2, code #1., etc.
[THIS PART CORRESPONDS TO ASSIGNMENT #3. NOTE THE INDIVIDUAL REUIREMENTS PART.]
1. MATLAB LISTING FOR COMPUTING THE EIGENFEATURE VECTOR FOR A GIVEN FACE IMAGE. [Section 3, code #1, etc.; at the end of the report]
2. MATLAB LISTING FOR RECONSTRUCTING A FACE IMAGE GIVEN THE EIGENBASIS AND OTHER RELEVANT PARAMETERS.
3. INDIVIDUAL PART: EACH OF THE MEMBERS SHOULD SUBMIT. PLEASE IDENTIFY YOUR NAME CLEARLY.
In this part, you compute the feature vector and the reconstruct your face image using the first 20, 40, and 60 eigenfaces. You print out the following in your report: your original image and your reconstructed image, for each of the facial expressions. Organize the pictures in a tabular format as follows:
| original | reconst (20) | reconst (40) | reconst (60) | |
|---|---|---|---|---|
| normal I | ||||
| normal II | ||||
| happy | ||||
| angry | ||||
| surprised | ||||
| disgusted |
[THIS PART CORRESPONDS TO ASSIGNMENT #4. HERE YOU WILL PROVIDE RESULTS ON FACE RECOGNITION. THERE IS AN INDIVIDUAL REQUIREMENTS PART AS WELL.]
1. MATLAB LISTING FOR RETRIEVING THE FIVE CLOSEST FACE IMAGES IN THE DATABASE GIVEN AN INPUT QUERY FACE. ASSUME EUCLIDEAN DISTANCES IN THE FEATURE SPACE.
2. INDIVIDUAL PART: PERFORMANCE OF THE METHOD WHEN THE QUERY IMAGE CORRESPONDS TO YOUR FACE. Show the first five best retrievals using 10 eigenfaces as the basis. If your face is not in the first best retrieval, keep increasing the size of the feature vector till you get your face in the top spot. Do this for each of the different expressions. Just record the number of feature dimensions, you do not have to print out the results other than for K=10. note that there is no guarantee that even with K=66 that your face will be in the first place. Make sure that you note this if you are not able to retrieve your face in the first position. Comment on the overall results.
3. FACE RECOGNITION ACCURACY: FILL IN THE TABLE. Here you will be computing the average recognition statistics over all the face images in the database. You will be computing what percentage of times the first retrieval is actually the correct one. Note that this measure may lead to very low recognition accuracy. Optional part: you may want to comment on what percentage of the times the correct retrieval is in the top 3 matches and in the top 5 matches.
| Normal I | Normal II | Happy | Angry | Surprised | Disgusted | |
|---|---|---|---|---|---|---|
| K=10 | ||||||
| K=30 | ||||||
| K=60 |
4. MATLAB CODE YOU DEVELOPED FOR COMPUTING THE RECOGNITION ACCURACY.
CONCLUDING REMARKS. ANY OBSERVATIONS ON THE EFFECTIVENESS OF THE EIGENFACE BASED APPROACH.