Belkasoft Evidence Center Help Contents
< Previous help topic
The product allows you
to detect faces in picture and video profiles. In order to detect face, select picture or video profile of interest, right click it, choose Analyze pictures context menu item and then Detect face. The analysis will start.
After the analysis the filter Pictures with faces under Analysis
results profile subnode in Case Explorer will be filled
with the pictures with faces detected. The total number of pictures with faces will
be indicated in the
4 pictures with faces detected
The analysis is designed to work with clear frontal and profile faces. The best results can be achieved on frontal faces. Other kinds of faces, e.g. turned halfway, might not be detected as there is no robust algorithm existing so far. The following obstacles may prevent successful face detection:
- Face significantly turned from frontal or profile position
- Something covers a major area on the face (e.g. sunglasses, scarf or hand)
- Poor lighting conditions
Face detection uses neural networks in order to find faces. False negative and false positive detections are to be expected. For example, the algorithm frequently incorrectly detects knees area as face. To reduce false positives, the product performs a number of additional types of analysis such as:
- Skin detection
- Nose detection
- Eyes detection
- Mouth detection
To get rid of false positives you can filter faces that only contain all of those features. Note: false positives and false negatives counts depend on each other. If you decrease the number of false positives, more false negatives will appear (so you may miss some pictures with faces).
To see which features are detected for a face, you can inspect the Analysis properties section in the Item Properties window:
For each face discovered, a separate record is created with all the details such as face bounds and additional features such as nose, mouth, skin and eyes detected.
Each face is surrounded with a green box inside Picture List and Picture Viewer windows, in case correspondent option is set in Picture Options. The product can successfully
detect not only real faces but also faces within a drawing or even a cartoon!
Note: for face extraction from video file, you should
extract key frames first.
< Previous help topic