3D Video Coding Using Just Noticeable Depth Difference Based on H.265/HEVC

Abstract

Just noticeable depth difference (JNDD) model represents the depth sensitivity of human visual system (HVS). Human eyes have different depth sensitivity for different distance and binocular disparity information. The JNDD model indicates that human can easily perceive the depth change when an object is close to the screen plane, but less sensitive to a distant object. If the depth information is below the JNDD threshold, there are visual redundancies, and we can process a video with bigger compression ratio. In this paper, we apply Silva's JNDD model to H.265/HEVC for the color image coding. We propose a perception region division algorithm (PRDA), through which we divide the depth map into four regions with different depth. Then, the quantization parameter is adjusted depending on a visual perception threshold. Experimental results show that our color image coding scheme improves the image quality without increasing bitrates.

A Survey of Stereoscopic 3D Just Noticeable Difference Models

“…Similarity, Ding et al [96] recently developed a depth map preprocessing method using JNDD to improve the 3D extension of the high efficiency video coding (HEVC) standard. Bai et al [97] applied JNDD in H.265/HEVC for color image coding by adjusting the quantization parameter (QP). JNDD has also been employed in depth sensation enhancement [98]- [100].…”