G-NetFace

About G-NetFace

G-NetFace uses neural network for face recognition. You can compare two faces or select face and search folder with images for face match.

COMPARE FACES:
Load two images and calculate similarity between faces. If an image contains several faces you can select which one to use for comparison.

How to use:
1. Load face 1 image.
2. Load face 2 image.
3. Press COMPARE FACES - similarity between faces is calculated. You can also see chart of the 128 face embeddings that neural network produces.

FOLDER SEARCH:
Load face image and search matching face in a folder of images.

How to use:
1. Load face image.
2. Select folder with images to be searched.
3. Press START - the selected face is compared with faces in images in selected folder and the results with similarity bigger than defined threshold in settings are shown.

DATABASE SEARCH:
How to use:
1. Create database - Open Menu - Database. Select images folder and database and press START. Images from selected folder are imported in selected database.
2. Open Menu - Database search. Select face image and database to search
3. Press START - the selected face is compared with faces in selected database and the results with similarity bigger than defined threshold in settings are shown.

Settings:
- Folder search threshold - defines threshold which results to appear in FOLDER SEARCH. Faces with similarity bigger than this threshold will be shown.
- Face detection area shrink percentage - control the size of the area around the face. Setting positive values will shrink the area and negative values will expand it.
- Face detection area lowering percentage - control the height of the area around the face. Setting positive values will lower the area around face and negative values will lift the area.
- Always align faces - when this is checked the faces will be automatically aligned. This will improve the results, but will slow the process.

By adjusting Face detection area shrink and lowering percentage you can control what part of the image is matched.