Binocular stereo matching is one of the core algorithms in stereo vision. The edge-aware filter-based cost aggregation methods can produce precise disparity maps on the famous Middlebury benchmark (indoor). However, they perform poorly on the KITTI vision benchmark (outdoor) because frontal-parallel surfaces are assumed in the filter-based methods. We propose a new cost aggregation algorithm which discards the frontal-parallel surfaces assumption. The proposed algorithm performs like optimizing an energy function via dynamic programming. The proposed energy function integrates the pairwise smooth energy by the edge-aware filtering approach, which makes the proposed method adapt to slanted surfaces. The proposed algorithm not only outperforms the edge-aware filter-based local methods on the Middlebury benchmark but also performs well on the KITTI vision benchmark.
We introduce a new superpixel segmentation algorithm in this paper with a real-time performance that make the practical in the machine vision systems. The algorithm is divided into two steps. First, a simple linear clustering with a O(N) complexity is used for efficient initial segmentation. Second, to further optimize the boundary localizations, a region competition skill is first used on the superpixels’ edge points and then iterates on the unstable edge points. As only the superpixels’ edge points are considered and most edge points become stable quickly, the clustering samples are significantly compressed to speed up the process. Experimental results on the Berkeley BSDS500 dataset show that the segmentation quality of the proposed method is slightly better than the SLIC algorithm, which is a state-of-the-art superpixel segmentation algorithm. In addition, the average speed achieves speedups of about 5X from the original SLIC algorithm, more than 30 frames per second to process 481x321 images in BSDS500.
We present a way to construct a complete set of scaling rotation and translation invariants extract directly from Zernike moments. Zernike moment can be constructed by Radial moment. In our method in order to construct invariant Zernike moment is to achieve invariant Radial moment which is component of Zernike moment. We use matrix form to denote relationship between Radial and Zernike moment, which makes derivation more comprehensible. The translation invariant Radial moment is first introduced, for it is most complicated part of all the three invariant. Rotation and scaling invariant Radial moment is achieved by normalizing the factor caused by rotation and scaling. The form of invariant radial moment is to combine three parts of invariant. Some experiment has done to test the performance of invariance. In this experiment we take an image library containing 23,329 files which are built by translation rotation and zoom in out of one origin Latin character image. Most of the value of standard deviation ratio by mean of proposed moments is nearly 1%. In addition, retrieval experiment is to test the discrimination ability. MPEG-7 CE shape1 - Part A library is taken in this experiment. The recall rate in part A1 is 96.6% and is 100% in part A2.