A Scheme for Face Recognition in Complex Environments

2. Related Work

Face detection in complex environments has achieved very remarkable results, such as the typical algorithms with long standing impact e.g. Adaboost, etc (Viola and Jones 2001) and Skin Color fusion model (Azad et al. 2015). However, most of the recent researches are focusing on Skin Color instead of background processing. Usually face recognition is able to be completed in four steps: 1) Locate a face based on image feature and template (such as eyes, nose and mouth); 2) Segment the face in the image (face shape based); 3) Normalize the frontal face; 4) Recognize the face with the eigenface method. However, the precision of face recognition has not been deeply taken into consideration yet due to various reasons.

Yuan et al. proposed a method to detect frontal faces in a complex environment by using image segmentation and face detection and verification. The face detection and verification are regarded as consisting of Valley-like detector and face verification with positive-negative attractor (Yuan et al. 2002). The Valley-like detector distinguished the differences between surrounding pixels and core ones in a mosaic image. The positive-negative is used to identify the positive attractor such as eyes and mouth, the negative attractor such as the surround pixel (Yuan et al. 2002). However, all of the previous studies is in focus of the face detection and recognition without filtering the expected and unexpected face.

The MOR is a convenient way of removing moving objects from a video by using frame differencing (Prabhakar et al. 2012). The frame differencing is the technique that could differentiate the variations of two adjacent video frames from both spatial and frequency domains and used them to distinguish whether an object has any motions and how fast or slow it moves (Prabhakar et al. 2012).

In this paper, our contribution is to minimize the possibility of false alarms triggered by unwanted faces being detected and recognized by the face recognition system. To our knowledge, this is the first time the issue has been discussed for this kind of applications.