Matching facial photographs to composite sketches.
The software retreive the photos closest to the composite sketch given in input.
Steps:
- Use Haar-like features to detect face and facial landmarks.
- Compute HOG and Sift descriptors of hair, brow, eyes, nose, mouth.
- Project descriptors to more discriminant space using LDA.
- Use euclidean distance of descriptors to sort photos.
- Final ranking is calculated by using Borda count
The first time you run a search it would take much time because the program has to build descriptors and projecting vectors. Descriptors and vectors will be saved respectively as descriptors.bin and lda.bin, both are saved in the bin directory.
If you want to save computational time, you can download pre-build descriptors and vectors from the release section of GitHub. Follow the instruction in the release.
-
Populate the database directory.
-
If you want to use your database:
Please check the paths and file extentions specified as constants in ImageRetreivalSystem.cs
-
If you want to use the suggested database:
The file database-link.txt contains links to suggested database. The sub-directories of UoM-SGFS-v2 have txt files containing the list of used files.
Please, note that only frontal face images without duplicates have been kept, so, rely on images list.
Dir Sketches/Set A and Sketches/Set B contains only selected images from UoM-SGFS.
Dir Photos/Images contains only selected images from FERET.
Dir Photos/Others contains selected images from FERET, MEDS-II, and FEI.
-
-
Run with Visual Studio
Required libraries in bin are already linked in project files.