I'll go a bit more into detail explaining :ocv:class:`FaceRecognizer`, because it doesn't look like a powerful interface at first sight. But: Every :ocv:class:`FaceRecognizer` is an :ocv:class:`Algorithm`, so you can easily get/set all model internals (if allowed by the implementation). :ocv:class:`Algorithm` is a relatively new OpenCV concept, which is available since the 2.4 release. I suggest you take a look at its description.
* So called “virtual constructor”. That is, each Algorithm derivative is registered at program start and you can get the list of registered algorithms and create instance of a particular algorithm by its name (see :ocv:func:`Algorithm::create`). If you plan to add your own algorithms, it is good practice to add a unique prefix to your algorithms to distinguish them from other algorithms.
* Setting/Retrieving algorithm parameters by name. If you used video capturing functionality from OpenCV highgui module, you are probably familar with :ocv:cfunc:`cvSetCaptureProperty`, :ocv:cfunc:`cvGetCaptureProperty`, :ocv:func:`VideoCapture::set` and :ocv:func:`VideoCapture::get`. :ocv:class:`Algorithm` provides similar method where instead of integer id's you specify the parameter names as text strings. See :ocv:func:`Algorithm::set` and :ocv:func:`Algorithm::get` for details.
* Reading and writing parameters from/to XML or YAML files. Every Algorithm derivative can store all its parameters and then read them back. There is no need to re-implement it each time.
Sometimes you run into the situation, when you want to apply a threshold on the prediction. A common scenario in face recognition is to tell, wether a face belongs to the training dataset or if it is unknown. You might wonder, why there's no public API in :ocv:class:`FaceRecognizer` to set the threshold for the prediction, but rest assured: It's supported. It just means there's no generic way in an abstract class to provide an interface for setting/getting the thresholds of *every possible*:ocv:class:`FaceRecognizer` algorithm. The appropriate place to set the thresholds is in the constructor of the specific :ocv:class:`FaceRecognizer` and since every :ocv:class:`FaceRecognizer` is a :ocv:class:`Algorithm` (see above), you can get/set the thresholds at runtime!
Here is an example of setting a threshold for the Eigenfaces method, when creating the model:
// create the concrete implementation with the appropiate parameters:
Ptr<FaceRecognizer> model = createEigenFaceRecognizer(num_components, threshold);
Sometimes it's impossible to train the model, just to experiment with threshold values. Thanks to :ocv:class:`Algorithm` it's possible to set internal model thresholds during runtime. Let's see how we would set/get the prediction for the Eigenface model, we've created above:
..code-block:: cpp
// The following line reads the threshold from the Eigenfaces model:
Adding new images to a trained :ocv:class:`FaceRecognizer` is possible, but only if the :ocv:class:`FaceRecognizer` supports it. For the Eigenfaces and Fisherfaces method each call to :ocv:func:`FaceRecognizer::train` empties the old model and estimates a new model on the given data. This is an algorithmic necessity for these two algorithms, no way around that. Please see the tutorial Guide To Face Recognition with OpenCV for details. If you call :ocv:func:`FaceRecognizer::train` on a LBPH model, the internal model is extended with the new samples.
Please note: A :ocv:class:`FaceRecognizer` does not store your training images (this would be very memory intense), the caller is responsible for maintaining the dataset.
Getting the name of a FaceRecognizer
+++++++++++++++++++++++++++++++++++++
Since every :ocv:class:`FaceRecognizer` is a :ocv:class:`Algorithm`, you can use :ocv:func:`Algorithm::name` to get the name of a :ocv:class:`FaceRecognizer`:
..code-block:: cpp
// Create a FaceRecognizer:
Ptr<FaceRecognizer> model = createEigenFaceRecognizer();
The following source code snippet shows you how to learn a Fisherfaces model on a given set of images. The images are read with :ocv:func:`imread` and pushed into a ``std::vector<Mat>``. The labels of each image are stored within a ``std::vector<int>`` (you could also use a :ocv:class:`Mat` of type `CV_32SC1`). Think of the label as the subject (the person) this image belongs to, so same subjects (persons) should have the same label. For the available :ocv:class:`FaceRecognizer` you don't have to pay any attention to the order of the labels, just make sure same persons have the same label:
Now that you have read some images, we can create a new :ocv:class:`FaceRecognizer`. In this example I'll create a Fisherfaces model and decide to keep all of the possible Fisherfaces:
..code-block:: cpp
// Create a new Fisherfaces model and retain all available Fisherfaces,
// this is the most common usage of this specific FaceRecognizer:
//
Ptr<FaceRecognizer> model = createFisherFaceRecognizer();
:param num_components:The number of components (read: Eigenfaces) kept for this Prinicpal Component Analysis. As a hint: There's no rule how many components (read: Eigenfaces) should be kept for good reconstruction capabilities. It is based on your input data, so experiment with the number. Keeping 80 components should almost always be sufficient.
* Training and prediction must be done on grayscale images, use :ocv:func:`cvtColor` to convert between the color spaces.
***THE EIGENFACES METHOD MAKES THE ASSUMPTION, THAT THE TRAINING AND TEST IMAGES ARE OF EQUAL SIZE.** (caps-lock, because I got so many mails asking for this). You have to make sure your input data has the correct shape, else a meaningful exception is thrown. Use :ocv:func:`resize` to resize the images.
:param num_components:The number of components (read: Fisherfaces) kept for this Linear Discriminant Analysis with the Fisherfaces criterion. It's useful to keep all components, that means the number of your classes ``c`` (read: subjects, persons you want to recognize). If you leave this at the default (``0``) or set it to a value less-equal ``0`` or greater ``(c-1)``, it will be set to the correct number ``(c-1)`` automatically.
:param threshold:The threshold applied in the prediction. If the distance to the nearest neighbor is larger than the threshold, this method returns -1.
* Training and prediction must be done on grayscale images, use :ocv:func:`cvtColor` to convert between the color spaces.
***THE FISHERFACES METHOD MAKES THE ASSUMPTION, THAT THE TRAINING AND TEST IMAGES ARE OF EQUAL SIZE.** (caps-lock, because I got so many mails asking for this). You have to make sure your input data has the correct shape, else a meaningful exception is thrown. Use :ocv:func:`resize` to resize the images.
:param radius:The radius used for building the Circular Local Binary Pattern. The greater the radius, the
:param neighbors:The number of sample points to build a Circular Local Binary Pattern from. An appropriate value is to use `` 8`` sample points. Keep in mind: the more sample points you include, the higher the computational cost.
:param grid_x:The number of cells in the horizontal direction, ``8`` is a common value used in publications. The more cells, the finer the grid, the higher the dimensionality of the resulting feature vector.
:param grid_y:The number of cells in the vertical direction, ``8`` is a common value used in publications. The more cells, the finer the grid, the higher the dimensionality of the resulting feature vector.
:param threshold:The threshold applied in the prediction. If the distance to the nearest neighbor is larger than the threshold, this method returns -1.
* The Circular Local Binary Patterns (used in training and prediction) expect the data given as grayscale images, use :ocv:func:`cvtColor` to convert between the color spaces.
