Applied Mechanics and Materials Vols. 519-520

Abstract: Abstract. This paper proposes an incomplete GEI gait recognition method based on Random Forests. There are numerous methods exist for git recognition,but they all lead to high dimensional feature spaces. To address the problem of high dimensional feature space, we propose the use of the Random Forest algorithm to rank features' importance . In order to efficiently search throughout subspaces, we apply a backward feature elimination search strategy.This demonstrate static areas of a GEI also contain useful information.Then, we project the selected feature to a low-dimensional feature subspace via the newly proposed two-dimensional locality preserving projections (2DLPP) method.Asa sequence,we further improve the discriminative power of the extracted features. Experimental results on the CASIA gait database demonstrate the effectiveness of the proposed method.

Abstract: Considering features of PCB marks and components image, in order to search a consistent location for reference images in multidimensional parameter space, the algorithm of geometric characteristic recognition location is proposed. Geometric location algorithm mainly include model training and real-time search. Compared to edge location and traditional localization methods, which not only adapt the gray-scale linearity and the gray non-linear changes, but also support changes in scale and perspective. Numerical results shows that the position deviation of geometric positioning algorithm is less than 0.5 pixel, the angle deviation is less than 0.5 degree. This algorithm is robust, simple, practical and it is better than the traditional location method.

Abstract: The quality of side information (SI) is one of the critical factors which play an important part in the compression performance of Distributed Video Coding (DVC).In this paper, we analyze the algorithm of Motion Compensation Interpolation (MCI) and propose an improving algorithm against its limitation. The algorithm proposed firstly makes each pixel has more than one motion vectors, then picks the best vector out and makes it plays the most important role in motion compensation with weighted average. Through the simulation of different frame sequences, the PSNR of the side information generated by the algorithm proposed in this paper is 0.3-0.7dB higher than that in the original algorithm.

Abstract: Shape from shading (SFS) is a classical and important problem in the domain of computer vision. This paper presents a new image irradiance equation for perspective SFS method to reconstruct the hybrid surfaces that have both diffuse reflection and specular reflection. The hybrid reflectance model composed of a linear combination of Oren-Nayar model and Ward model is used to express the hybrid surfaces. An imaging model incorporating near point light source, perspective camera projection and the hybrid reflectance model is established. Under this model, the image irradiance equation has been derived as a non-linear partial differential equation (PDE). The resulting PDE is associated with a static Hamilton-Jacobi (H-J) equation considering the boundary conditions. Thus, the image irradiance equation of hybrid surfaces can be solved further.

Abstract: In this paper, we propose a novel ship detection model based on multi-cue visual attention mechanism, which include two steps. Firstly, the model acquires salient candidate regions across entire scene by using a bottom-up visual attention cue. The cue is saliency analysis method based on visual contrast. Then, a analysis method based on the environment around the ship is used to distinguish between ship and harbor.The method dont use the shape, edge or other forms of features of the ship objects. The ship detection results prove our method can effectively concentrate on the objects with greater contrast and distinguish between ship and harbor.

Abstract: The existing object tracking method using covariance modeling is hard to reach the desired tracking performance when the deformation of moving target and illumination changes are drastic, we proposed a object tracking algorithm based on bilateral filtering. Firstly, the algorithm deals the image to be tracked with bilateral filtering, and extracts the needed features of filtered image to construct covariance matrix as tracking model. Secondly, under log-Euclidean Riemannian metric, we construct similarity measure for object covariance matrix and model updating strategy. Extensive experiments show that the proposed method has better adaptability for object deformation and illumination changes.

Abstract: Threshold technique is one of the important techniques in image segmentation. Various thresholding segmentation techniques such as histogram, grayscale expectations, Otsu, maximum entropy and iterative are studied and compared by using Matlab 7.0. Experimental results show that the iterative method can perform well and get a better result than the other thresholding segmentation methods.

Abstract: Estimating the head pose is still a unique challenge for computer vision system. Previous methods at solving this problem have often proposed solutions formulated in a classification setting. In this paper, we formulate pose estimation as a regression problem to achieve robustness. We propose to use gradient orientation histograms based random regression forests for the task. Firstly, each sample image is divided into overlapped patches, and direction-sensitive features of patches are extracted. Then we train a random regression forest on these patches. Experiments are carried out on public available database, and the result shows that the proposed algorithm outperforms some other approaches in both accuracy and computational efficiency.

Abstract: A triangle rasterizer features the barycentric arithmetic for rasterization operation in graphics processors is introduced in this paper. First, the common method for triangle rasterization is presented, and its drawbacks of lacking simplicity and introducing accumutive errors are highlighted. Then, the new approach using the barycentric arithmetic is discussed in detail, including finding the pixels belong to the triangle and interpolating attributes for these pixels. At the basis of that, the hardware structure of the rasterizer is proposed with elaboration of the bounding rectangle establishment and the pipelined divider in it. At last, the RTL simulation result of the proposed rasterizer is given and analyzed, which has shown that the rasterizer can generate pixels efficiently and accurately.

Abstract: An automatic individual tree detection method from pure image is proposed. Color and texture features are selected to form a vector for a pixel-level classification, then trained to assign a label to each pixel. Other features can be integrated into the pixel vector for extracting more information of trees. An ensemble method combining multiple logistic regression classifiers improves effectiveness of the single pixel-level classifier. Then spectral, shape and knowledge characteristics of individual tree crowns are used for tree top localization. At last, tree crowns are delineated by region-based algorithm.