Face detection for android smartphone applications

Abstract

The purpose of this project is to provide a faster and more accurate means of utilizing camera related smart phone applications whose current trend might be cumbersome to individuals. The problem arises from the lack of sense of direction where a high-definition smart phone camera is pointing, due to its position being at the opposite side of the screen. In the frantic lifestyle of today, users want more than a trial-and-error method of taking self pictures. The solution would therefore be to interact minimally with the phone as much as possible by taking the picture correctly the first time. The application of Intelligent Computer System methods will minimize human interaction with the application, thus saving time, repetition and preserving correctness. The system proposed in this project makes use of a face detection system to identify any faces, and guide the user to place the detections as much as possible to the centre of the camera's field-of-vision using the smart phone speaker. The photo will be automatically captured once the detections are calculated to be close enough to the centre of the screen. A neural-network based face detection system is trained using 6000 training examples, but deemed too much false positives during evaluation to be implemented and be utilised in the application. An existent image-related Java library (JJIL) is then applied, making use of its Haar-Cascade based face detection system. This takes a few seconds to detect a face on typical Android smart phone processing capabilities, and therefore not deemed suitable for a real-time application need. The final application consists of Android's own face detection library, since this is optimised for use on devices with limited computational speed. This system is combined with the camera input feed to detect faces in real-time. The evaluation on the latter system was a success.