Improved block matching approach to fast disparity estimation

An improved block matching approach to fast disparity estimation in machine vision applications is proposed, where the matching criterion is the sum of the absolute difference（SAD）.By evaluating the lower bounds, which become increasingly tighter for the matching criteria, the method tries to successively terminate unnecessary computations of the matching criteria between the reference block in one image and the ineligible candidate blocks in another image.It also eliminates the ineligible blocks as early as possible, while ensuring the optimal disparity of each pixel.Also, the proposed method can further speed up the elimination of ineligible candidate blocks by efficiently using the continuous constraint of disparity to predict the initial disparity of each pixel.The performance of the new algorithm is evaluated by carrying out a theoretical analysis, and by comparing its performance with the disparity estimation method based on the standard block matching.Simulated results demonstrate that the proposed algorithm achieves a computational cost reduction of over 50.5% in comparision with the standard block matching method.

Hybrid tree guided PatchMatch and quantizing acceleration for multiple views disparity estimation

Existing stereo matching methods cannot guarantee both the computational accuracy and efficiency for ihe disparity estimation of large-scale or multi-view images.Hybrid tree method can obtain a disparity estimation fast with relatively low accuracy,while PatchMatch can give high-precision disparity value with relatively high computational cost.In this work,we propose the Hybrid Tree Guided PatchMatch which can calculate the disparity fast and accurate.Firstly,an initial disparity map is estimated by employing hybrid tree cost aggregation,which is used to constrain the label searching range of the PatchMatch.Furthermore,a reliable normal searching range for each current normal vector defined on the initial disparity map is calculated to refine the PatchMatch.Finally,an effective quantizing acceleration strategy is designed to decrease the matching computational cost of continuous disparity.Experimental results demonstrate that the disparity estimation based on our algorithm is better in binocular image benchmarks such as Middlebury and KITTI.We also provide the disparity estimation results for multi-view stereo in real scenes.

Performance Analysis of Disparity for Stereoscopic Image Pairs

Disparity is the geometrical difference between images of a stereoscopic pair. In this paper we give a comprehensive analysis of the statistical characteristics of disparity. Based on experiments, we discuss the relations between disparity, depth and object relation between block size and disparity estimation, and the influence of error criteria on disparity estimation.

Intermediate view synthesis from stereoscopic images

A new method is proposed for synthesizing intermediate views from a pair of stereoscopic images. In order to synthesize high-quality intermediate views, the block matching method together with a simplified multi-window technique and dynamic programming is used in the process of disparity estimation. Then occlusion detection is performed to locate occluded regions and their disparities are compensated. After the projecton of the left-to-right and right-to-left disparities onto the intermediate image, intermediate view is synthesized considering occluded regions. Experimental results show that our synthesis method can obtain intermediate views with higher quality.

Fast view prediction for stereo images based on Delaunay triangular mesh model

The view prediction is an important step in stereo/multiview video coding, wherein, disparity estil mation （DE） is a key and difficult operation. DE algorithms usually require enormous computing power. A fast DE algorithm based on Delaunay triangulation （DT） is proposed. First, a flexible and content adaptive DT mesh is established on a target frame by an iterative split-merge algorithm. Second, DE on DT nodes are performed in a three-stage algorithm, which gives the majority of nodes a good estimate of the disparity vectors （DV）, by removing unreliable nodes due to occlusion, and forcing the minority of ＇problematic nodes＇ to be searched again, within their umbrella-shaped polygon, to the best. Finally, the target view is predicted by using affine transformation. Experimental results show that the proposed algorithm can give a satisfactory DE with less computational cost.

Adaptive early termination of fast mode decision in MVC based on inter-view correlation

Variable size motion estimation （ME） and disparity estimation （DE） are employed to select the best coding mode for each macroblock （MB） in the current joint multiview video model （JMVM）. This technique achieves the highest possible coding efficiency, but it results in extremely large computation complexity which obstructs the multiview video coding （MVC） from practical application. This paper proposes an adaptive early termination of fast mode decision algorithm for MVC. It makes use of the coding information of the corresponding MBs in neighbor view based on inter-view correlation to early terminate the mode decision procedure. Experimental results show that the proposed fast mode decision algorithm can achieve computational 50% computation saving with no significant loss of rate distortion （RD） performance.

A three-dimensional measurement method for binocular endoscopes based on deep learning

In the practice of clinical endoscopy,the precise estimation of the lesion size is quite significant for diagnosis.In this paper,we propose a three-dimensional(3D)measurement method for binocular endoscopes based on deep learning,which can overcome the poor robustness of the traditional binocular matching algorithm in texture-less areas.A simulated binocular image dataset is created from the target 3D data obtained by a 3D scanner and the binocular camera is simulated by 3D rendering software to train a disparity estimation model for 3D measurement.The experimental results demonstrate that,compared with the traditional binocular matching algorithm,the proposed method improves the accuracy and disparity map generation speed by 48.9%and 90.5%,respectively.This can provide more accurate and reliable lesion size and improve the efficiency of endoscopic diagnosis.