Face Recognition Tool


tool_image

Description

It is an overview of HMMDemo application. See also hypertext manual on it.

This face recognition tool gives the computer the ability to recognize a face from image files (BMP format) or directly from an USB camera. The system requirements for this application are a Pentium® processor-based computer and Microsoft* Windows* NT 4.0, 98, or 2000. The face recognition uses the embedded hidden Markov model and implements the method described in "Face recognition using an embedded HMM" .

Controls

[1] Opens a new document.
[2] Opens an existing database or an image file in BMP format.
[3] Creates a new database. It gives the user the ability to specifty the file location where the information is to be stored.
[4] Adds a face to the training set. The rectangle in the right panel should bound the user's face. If the rectangle is misplaced or does not exist, create one by left-click and dragging a rectangle. If the left panel is not focused on a particular user, a dialog box will appear. In this box enter the name of the captured face. If the left panel is focused (by double-clicking a face) on a person, the captured face will automatically be added to the current set.
[5] Deletes an image from the training set or an entire set of images.
[6] Zooms in the images in the database (left panel).
[7] Zooms out the images in the database (left panel).
[8] Toggles video capture.
[9] Specifies the parameters of the video, the resolution of the image, the depth and size of the video frames.
[10] Launches the capture device properties for your USB camera.
[11] Train the face model. If the left panel is not focused on a set, all sets will be trained. If the left panel is focused on a set, the current set is trained.
[12] Run recognition algorithm on image area within bounding box in the right panel. If there is no bounding box, create one by left-click and dragging a rectangle.
[13] Display about box.
[14] Not enabled

How to...

Create a new face database

Click button [3] on the toolbar to create the file where the database information is stored. Specify the filename and database description in the dialog box.
To add images from camera or file to the face database, specify the rectanglular area bounding the face by left-click and dragging. Then click button [4] to add the face to the database. In the dialog box, enter the name of the captured face. The left panel will automatically focus on the set for this particular person. Incorrect images can be removed by cselecting the image and clicking button [5]. To restart the video after a face image was added to the database, click button [8]. Use the File menu, option Save to save the database file.

Open an existing face database

Click button [2] on the toolbar to specify the file where the database information is stored. One image for each person in the database will be displayed on the left panel. To view all the images of a particular person, double-click the specific face. All the images in the database representing the specified person are displayed in the left panel. To return to the previous screen, double-click on any white space in the left panel.

Train a face model

After capturing several instances for a person, train the face model for that person by clicking button [10]. The models can be trained one by one or all at a once. To train one person, double-click the person to be trained and click button [10]. To train everyone, make sure the left panel is not focused by double-clicking any whitespace in the left panel and click button [10]. Note that when an image is added or deleted from a particular set, that set needs to be retrained.

Recognize from live video or from file on disk

If all the people in the database are trained, faces can be recognized. To recognize a face from a camera view or from a file, first create a bounding box in the right panel by left-click and dragging. Then click button [12] to run the recognition algorithm. The recognition results box will list in order of likelihood the three people most likely to match the image. The image in the left panel corresponding to the most likely person will also be shown in a with a colored border. The test image can also be added to the database by clicking button [4]. In this case, the corresponding model must be retrained.


Back to Top