Face Detection in MemoriesOnTV 4.

10th December | Tutorial

One of the exciting features in MemoriesOnTV 4 is the availability of automatic face detection when doing pan/zoom. With this, it is trivial to pan to each individual face in a group shot.

Let’s take a quick look at how this works. Let’s import a group shot into MemoriesOnTV, and visit its Pan/Zoom setting:

When you select “Detect face and pan to each other”, MemoriesOnTV will automatically scan your picture for upright frontal faces. It will create a key frame at each of the detected faces, such that the camera pans through all of them with a slight pause at each.

It is important to note that the face detection works for upright frontal faces. If you have a face that’s facing too much sideway, it will not be recognized. The face detection algorithm has a good accuracy of > 95%, so it will work well without further tweaking most of the time.

There are instances whereby the face detection doesn’t detect a face (eg. a face that’s not upright), or that it detects a non-face as one. You can address such situations quite simply :

Missing Faces:

For faces that aren’t detected, you can manually add them by adding a new key frame that is zoomed into the missed face. In the following picture, the face detection detects all but 1 face. To add the undetected face, we add a new key frame (via “menu -> Insert New Frame”) between its previous & next faces:

We now have all the faces, but we will still need to adjust the key frame timings evenly so that the panning speed is equal for each face. To do this, we select “Distribute Frames Evenly” from the “menu”. This will set each key frame to a time that’s evenly distributed.

Wrong Detection:

The face detection algorithm will sometimes detect wrongly, and recognizes a non-face as one. For such cases, we will need to delete away the wrong key frame. Select the key frame that shows the wrongly detected face, and then select “menu -> Remove Frame”.

After this, we will need to re-distribute the remaining key frames, so that they are evenly spreaded on the timeline. For this, we select “menu -> Distribute Frames Evenly”.

Face Detection Sensitivity

If you have a group shot with many people (eg. > 20), it is likely that each face on the photo will be very small. This can possibly escape face detection.
Assuming your photo is of sufficient resolution, you can still get face detection to work correctly on such photos, by increasing its sensitivity in the Preference window (”Edit -> Preference” menu):

Do note that increased sensitivity will lead to increased false positives as well, so you want to avoid too sensitive a setting. In general, the default sensitivity will work well for a wide range of group shots.

Powered by WordPress