Face Crop uses computer vision to automatically detect faces in images and applies a rectangular crop focused on either all faces or on the biggest face.
- Deep Neural Network algorithm. Select deep neural network for improved accuracy and to control the level of confidence in the face detection before applying the algorithm. When you select deep neural network, faces only need to be discernible when the image is scaled up or down to 300x300 pixels to be eligible for detection.
- Cascade Classifier algorithm. If you
select the cascade classifier algorithm you need to make sure that the faces you are
detecting are at least four percent the size of the largest dimension (height or width) of
the image. To be eligible for detection, faces also need to be at least at least 20x20
pixels in size.
For example, if a face is 50x50 pixels in size and the image width is 1700 pixels the cascade classifier algorithm will not accurately perform the face crop. Four percent of 1700 is 68, so this face is too small in relation to the image width.
If however, the image width is only 1000 pixels the cascade classifier algorithm can successfully perform the face crop. The face is more than four percent of the image width.
In most cases, it is sufficient to set the focus and padding and to accept the defaults for other fields for this transformation. The maximum and minimum size options scale the cropped image to the width and height you specify. The scaled result has the same aspect ratio as the crop dimensions. If you set only one dimension (width or height), Image and Video Manager scales the cropped area to that size, maintaining the aspect ratio of the region of interest.
If you add the IMQuery transformation to a policy, use the im variable, and select Face Crop, you can use a query string to automatically crop any face in an image using this policy. See Syntax and Examples for the syntax for the query string parameter.