FaceSDK 4.0 Release Notes

FaceSDK 4.0 improves the speed and accuracy of facial feature detection, with 66 facial features detected in real-time. The 4.0 version supports multi-core processors to boost the performance of face recognition, face detection, and facial feature detection. Support for MJPEG IP cameras has been added.

The increase in speed when using all cores of an Intel i7 processor is 315% for facial feature detection,^* 250% for face detection, 130% for face template creation, and 225% for eye detection. The facial feature detection speed is 9.3 frames per second (including face detection using real-time parameters).

Support for Java language has been added. A CImage class for easy manipulation with images on .NET has been added. JPEG file loading was speeded up by 390%.^**

More samples were added:

• Lookalikes samples (C#-DB, C#-SQLite, C-DB, C-SQLite) to work with Microsoft SQL and SQLite databases (C#.NET, C++)
• LiveFacialFeatures to detect facial features in real-time (C#.NET, VB.NET, Java, Delphi, C++)
• IPCamera to work with IP cameras (C#.NET, VB.NET, Java, C++)
• FaceTracking, FacialFeatures, and LiveRecognition now feature a Java version.

The LiveRecognition demo application (available on the Welcome screen after the installation) introduces new, more robust usage of FaceSDK recognition algorithms. The Facial Feature Demo application, which tracks a person's facial features in real-time, was added.

Functions to control the number of cores used by FaceSDK were added.

Unicode filename support on Windows was added.

Working with multiple cameras is now thread-safe.

More technical details on migrating from version 3.0 to 4.0 are available in the documentation.

 

---

  ^* The speed is measured on an Intel Core i7 930 processor using 8 threads, compared to the same processor using 1 thread.
^** The speed is measured on a collection of 1,000 random JPEG files of various resolutions.

 

 

FaceSDK 3.0 Release Notes

FaceSDK 3.0 introduces much faster and reliable face recognition. The speed of creating templates for faces was improved by 640% (from 1.3 frames per second to 8.5 frames per second).^* The rate of person’s acceptance when lighting conditions vary was improved by 5 times (from 65% to 93%).^** This allows to recognize a person in real time regardless of illumination, be it daylight or artificial. Version 3.0 also introduces real-time detection of eye centers.

More samples were added for C# .NET, VB.NET, C/C++, VB6, Delphi and Borland C++ Builder:

• LiveRecognition remembers a person and recognizes her in real time (using a webcam)
• FaceTracking tracks faces in real time
• Lookalikes creates the database of faces and searches for best matches
• FacialFeatures detects facial features on a photo
• Portrait crops a face in command line

Numerous demo applications were added, available in the Welcome screen right after the installation:

• Webcam demo remembers and recognizes persons from a webcam
• Photo demo detects faces on photos
• Panorama demo displays a 3D panorama, tracking the user’s head to change the point of view

The integration with .NET becomes simpler with new FaceSDK.NET assembly. The assembly does not require facesdk.dll and facesdkcam.dll which are already contained inside the assembly.

The facesdkcam.dll file was merged with facesdk.dll, simplifying the deployment and version control for C/C++, Delphi, VB6 and Borland C++ Builder developers.

License Key Wizard was added to simplify requesting an evaluation key from Luxand and pasting it into samples.

Face matching returns a more intuitive value of similarity – a value close to the probability that templates belong to the same person.

The template size was reduced from 92 kilobytes to 16 kilobytes.

Now a camera can be distinguished by its unique ID (device path), which is helpful when multiple cameras of the same manufacturer are used.

A function to mirror the image received from camera was added.

More technical details on migration from version 2.0 to 3.0 are available in the documentation.

 

---

   ^* Measured on Intel 2.4 Ghz Processor
^** False rejection rate was lowered from 35% to 7% on FERET tests with typical false acceptance rate value = 1%.