Qingqing Yang

Full-Image Guided Filtering for Fast Stereo Matching

A novel full-image guided filtering method is proposed. Different with many existing neighborhood filters, all input elements are employed during the proposed filtering approach. In addition, a novel scheme called weight propagation is proposed to compute support weights. It fulfills the requirements of edge preserving and low complexity. It is applied to the cost-volume filtering in the local stereo matching framework. The algorithm utilizing the proposed filtering method is currently one of the best local algorithms on the Middlebury stereo testbed in terms of both speed and accuracy.

Depth-image-based rendering with spatial and temporal texture synthesis for 3DTV

A depth-image-based rendering (DIBR) method with spatial and temporal texture synthesis is presented in this article. Theoretically, the DIBR algorithm can be used to generate arbitrary virtual views of the same scene in a three-dimensional television system. But the disoccluded area, which is occluded in the original views and becomes visible in the virtual views, makes it very difficult to obtain high image quality in the extrapolated views. The proposed view synthesis method combines the temporally stationary scene information extracted from the input video and spatial texture in the current frame to fill the disoccluded areas in the virtual views. Firstly, the current texture image and a stationary scene image, which is extracted from the input video, are warped to the same virtual perspective position by the DIBR method. Then, the two virtual images are merged together to reduce the hole regions and maintain the temporal consistency of these areas. Finally, an oriented exemplar-based inpainting method is utilized to eliminate the remaining holes. Experimental results are shown to demonstrate the performance and advantage of the proposed method compared with other view synthesis methods.

Hierarchical Joint Bilateral Filtering for Depth Post-Processing

Various 3D applications require accurate and smooth depth map, and post-processing is necessary for depth map directly generated by different correspondence algorithms. A hierarchical joint bilateral filtering method is proposed to improve the coarse depth map. By first carrying out depth confidence measuring, pixels are put into different categories according to their matching confidence. Then the initial coarse depth map is down-sampled together with the corresponding confidence map. Depth map is progressively fixed during multistep up sampling. Different from many filtering approaches, confident matches are propagated to unconfident regions by suppressing outliers in a hierarchical structure. Experiment results present that the proposed method can achieve significant improvement of initial depth map with low computational complexity.

Fast local stereo matching using two-level adaptive cost filtering

Recent local stereo correspondence algorithms achieve accurate results by performing effective cost aggregation. In this paper, we solve the cost aggregation problem in the view of cost-volume filtering. A novel concept named “two-level local adaptation” is introduced to guide the proposed filtering approach. Also, a novel post-processing method is proposed to handle both occlusions and textureless regions. The improvement of performance is confirmed by applying it to the proposed stereo correspondence algorithm. The overall method generates competitive results, and outperforms methods that use the similar filtering technique. By implementing the entire algorithm on the GPU, it can achieve about 10 frames/s for typical stereo pairs with a resolution of 640×360 and a disparity range of 20 pixels.

A novel guided image filter using orthogonal geodesic distance weight

In many applications, meaningful image structures tend to be piecewise smooth rather than band-limited. And edge-preserving image filters are always used to extract such structures and reduce noise. In this paper, a new image filtering method is introduced. We use geodesic distance to compute support weight, and a fast approximation named orthogonal geodesic distance weight is proposed. It greatly reduces the computational complexity while keeping excellent performance. The proposed filter is especially suitable for filtering with a high quality guide image.

Multiview virtual image synthesis for auto-stereoscopic display based on two views

In this paper, an efficient depth image based view rendering method for high quality multiview virtual images synthesis is proposed, which utilizes two color images with their associated depth maps. The proposed method consists of four main steps. The first step is virtual camera parameters design by using shift-sensor camera model to avoid the keystone distortion. Secondly, the mixed color pixels around the boundary of image objects is detected and skipped in the following rendering step, which can eliminate the “ghost” artifacts. Then the two input color images are rendered to multiple virtual view positions on the basis of depth-image-based rendering (DIBR) and merged into one image. Finally, an oriented exemplar-based inpainting algorithm is proposed to fill the disoccluded areas. Experiment results are shown to demonstrate the superior performance of the proposed method compared to other traditional DIBR methods.

Hybrid stereo matching by dynamic programming with enhanced cost entry for real-time depth generation

In this paper, a hybrid stereo matching method combining cross-based adaptive window aggregation and basic dynamic programming (DP) is proposed. A well-known drawback of DP is the streaking artifact, which greatly influences the subjective effect. Rather than introducing variable tree structures, we tackle this problem by providing an enhanced cost entry for DP optimization. Inter-scanlines are related closely by cost aggregating within a surface adaptive window, in which way the artifact is greatly reduced. Another contribution of our paper is that a variable starting point scheme is presented to force DP optimization to start from the accurate point, which prevents error cost accumulating at the beginning. Evaluation results show that our method outperforms most DP based algorithms. More importantly, the proposed hybrid method can be used in realtime systems. CUDA implementation of the proposed method can generate a disparity map in 15.17ms for a typical stereo pair with resolution of 384 × 288 and 16 disparity levels.