Eye-to-Eye Environment Creating Technique For Multi-View Video Conference

In multi-view video conference, eye-to-eye contact is important for effective interpersonal communication, it has some well documented functions in enhancing the feeling of friendliness, regulating conversation and improving the efficiency of capturing information. The method of creating eye-to-eye environment is introduced for the maintenance of eye-to-eye contact. It plays an important role in many fields, such as multi-view video conference, video phone and so on. Thus, it is very significant to study on the eye-to-eye environment creating technique.It can be usually supposed that listeners are directly faced to displaying screen in multi-view video conference. The creating of eye-to-eye environment means to recognize which camera image is frontal-view face image of the speaker from multiple camera images in multi-view video conference, where multiple cameras are arranged to capture the same conference scene. In this paper, a new method to create eye-to-eye environment is presented. The method is classified two steps, in the first step, face region is detected in each camera image and facial features are determined in the detected face region. A new algorithm of eye detection is proposed. A rectangle window for detecting eye is determined firstly, and then eye is detected according to a new criterion function, in which following factors are integratedly considered, the expectation of gray levels, the expectation of the variance of gray levels, the expectation of gradients of the gray levels and chroma information in the window. In the second step, the frontal-view face image of speaker is determined from multiple camera images according to the location of facial features relative to face region. The experimental results demonstrate that the proposed algorithms can detect face and facial features in real-time, and correctly judge which camera image is the frontal-view face image, so that eye-to-eye environment can be provided.