Reduced complexity multi-view video coding scheme for 2D camera arrays

As the number of views comprised in multi-view videos increases, some challenging problems emerge. Besides the bandwidth problems caused by the huge data flow, the calculation power needed by the multi-view encoder is an even higher burden than that of a single view encoder. In this paper, a complexity efficient way to encode a single-time instant of 5×3 view frames is presented. Some of the P frames known from the traditional encoding schemes have been replaced by a new type of frame called the D frame, in which the disparity vector of a block in a view can be derived from the other views due to the strong geometrical correspondence existing between adjacent views. Experimental results show that 20.2% complexity gain is achieved without compromising quality and bit-rate by wisely selecting threshold values at different QPs.

1. INTRODUCTION

In recent years, there has been a growing interest in multiview three-dimensional (3D) imaging systems and free viewpoint video (FVV) [1]–[3]. Basically, these imaging systems are capable of showing many view images captured from a scene through generated corresponding viewing zones. The viewer can see different viewpoints of the scene by changing his location from one viewing zone to the other. It is plausible that more viewing zones results in a more realistic experience. Apart from a few exceptions [4], most of the 3D imaging systems provide only horizontal parallax views due to the challenging problems such as complex optical design and implementation, high cost of the system and increased amount of image data [5], [6]. However, vertical views provide supplementary information that is indispensible for real 3D experience. Fig. 1. Prediction schemes proposed by Merkle et al. for views; (a) = MVCS-I, (b) = MVCS-2