Research On Down/Up Sampling Algorithm Of Depth Map In 3D Video

With the development of technology,3D video has become a new emerging trend in multimedia field.Comparing with traditional 2D video,3D video provides a better sense of reality and immersion because it has the depth information of the observed scenes. Depth sensor is the commonly technical means to obtain the depth information, which can capture the depth information timely and dynamically. Although the depth sensor is cheap and convenient, with the limit of the technology and the interference of noise, the depth image it captured usually has low resolution. In contrast, color images obtained from conventional video cameras have higher resolution. For the view synthesis purpose, depth image resolution should be the same as color image. Therefore, an efficient depth upsampler is necessary to convert depth image resolution from low to high. Besides depth maps generally have more spatial redundancy than natural images, this property can be exploited to compress a down-sampled depth image at the encoder which can reduce the amount of data transferred, then we upsample the down-sampled depth image to high resolution at the decoder. For better visual impact of 3D video, the accuracy of the depth map is the key, in this paper, we mainly focus on the depth map upsampling technology.First of all, we propose an effective adaptive depth sampling method with the saliency map. In the conventional down-up sampling depth encoding frame, the loss of depth boundary information in the processing of down sampling mainly affects the quality of view synthesis,because in the uniform sampling way we treat the edge and non-edge area fairly.in this paper, we take full advantage of the characteristics of depth map, relatively increasing the sampling points at the edge of the area can be greatly improve the quality of the decoding on the sampling depth map, reduce the sampling points in the smoothing area with saliency map. This method can improve the depth map accuracy near object boundaries.Then, we propose an SSIM-based improved joint bilateral upsampling method.JBU refers to color data under the assumption that depth edges usually correspond to color edges, it uses the pixel value and distance Gaussian filtering. However, visual artifacts are observed when the assumption fails. In this paper, we use the Structure Similarity Index Measurement as a weight to reduce the weight of the pixels which have the similar color value and different depth. In this way, the proposed upsampler generates high-resolution depth images preserving edge data.