Nostril Detection for Robust Mouth Tracking

author: Luca Cappelletta

Aim of this page is to better explain the "Nostril Detection for Robust Mouth Tracking" paper, submitted to Irish Signals and Systems Conference (ISSC) 2010.

The system described is a semi-automatic mouth region extractor, based on nostril detection and tracking.

Nostril Detection

The nostril detection process assumes that the nostrils are the darkest spots in the middle of the face area. Using this assumption and using a data refinement stage it is possible to track the nostril position during the video.

Figure 1: Nostril Detection.

Accumulated Difference Image - ADI

Using the nostril position a motion compensated Accumulated Difference Image (ADI) is generated. Result of this process is a logical mask corresponding of the mouth region. This mask is applied to every video frames, changing its position according to the nostrils position in each frame.

Figure 2: ADI.


The system has the ability to recover the nostril position in case of temporary failure or non-correct initialization.

Figure 3: Nostril recovery from temporary failure.
Figure 4: Nostril Recovery from non-correct initialization.
Page last modified on May 20, 2010