Depth map is an image that contains information relating to the distance of the surface of scene objects from a view point.
Deep convolutional grid warping network for joint depth map upsampling
In this work, we propose a novel deep learning-based scheme to upsample a depth map. Conventional deep learning-based scheme tried to modify pixel values using high-resolution color image and low-resolution depth map. Unlike these methods, the proposed method tries to solve depth upsampling problem by pixel position shift. We implement this idea using two sub-networks. One is the displacement network which determines pixel shifting vectors. The other is fusion network which shifts pixels and reconstructs high-resolution depth map.
[Y. Yang, D. Kim, B. T. Oh, “Deep convolutional grid warping network for joint depth map upsampling,” IEEE Access, Aug 2020.]
Depth map upsampling with a confidence-based joint guided filter
Joint depth map upsampling is a problem which enhances the resolution of an image using color image. One of the difficult problems is to find accurate position of edges. Recently, onion-peel filtering method is proposed which removes pixels in unreliable region first and fills the removed pixels sequentially in unreliable region from outside to inside. But, the performance of this method depends on the order in which the pixels are filled. Focusing on this, we propose novel filtering method which find optimal order using the confidence map derived from the shape of unreliable region, depth and color pixel value.
[Y. Yang, H. S. Lee, and B. T. Oh, “Depth map upsampling with a confidence-based joint guided filter,” Signal Processing: Image Communication, Sep. 2019.]
Pre-/Post-processing to improve the coding performance of multiview plus depth map
In this work, we propose new system for improving coding performance of multi-view and depth map (MVD) system using properties of depth map. The proposed method consists of two stages. In the pre-processing stage, dept map is filtered selectively to reduce the bit-rate cost while minimizing quality degradation. In the post-processing stage, decoded depth map is restored.
[D. Lee, Y. Yang, and B. T. Oh, “Pre-/post-processing to improve the coding performance of multiview plus depth map,” Signal Processing: Image Communication, Oct. 2017.]
[D. Lee, Y. Yang, and B. T. Oh,“Enhancement of depth map post-processing for 3D video coding,” International Workshop on Advanced Image Technology, 2016.]
Boundary artifacts reduction in view synthesis of 3D video system
In this work, we propose boundary artifacts reduction method for synthesized view in 3D video system. In this method, we analysis the property of boundary artifacts caused by compression noise. And, we set the convex set in pixel and frequency domain. Then, we find sub-optimal solution using projection onto convex sets method.
[이도훈, 양윤모, 오병태, “3차원 비디오의 합성영상 경계 잡음 제거,” 방송공학회논문지, Nov. 2016.]
[이도훈, 양윤모, 오병태, “3차원 비디오의 합성영상 경계 잡음 제거,” 한국방송미디어공학회 하계학술대회, 2016.]
[이도훈, 양윤모, 오병태, “깊이영상 해상도 조절에 따른 3차원 비디오 부호화 성능 분석,” 한국방송공학회 하계학술대회, 2015.]
[이도훈, 오병태, “다시점 영상 부호화 효율 향상을 위한 전처리 및 후처리 기법,” 영상처리 및 이해에 관한 워크샵, 2015.]