In this paper, we devise a novel steganography algorithm that has a high capacity while still retaining the ability of adjusting the embedding distortion. A shifting strategy is explored to embed the secret data into a given 3-D model effectively. In order to reduce the embedding distortion, we propose a truncated space of data instead of directly working on the critical geometric information from vertices of the cover model. The truncated space confines the distortion of each component of the stego-model within the space that means the embedding distortion could be controlled within a very low threshold. In theory, we can set the length of data truncation to adjust the embedding distortion below a specified level, at the cost of losing certain embedding capacity. Moreover, the embedding capacity is irrelevant to the shape of models, and the quality of the stego-model is mostly dependent on the length of truncation rather than the quantity of embedded secret data. The proposed 3-D steganography method has the capability to control the level of embedding distortion, and at the same time, has a high embedding capacity. Various experiments have demonstrated the flexibility and high performance of our new approach.

Skeleton-based action recognition is crucial for natural human-computer interaction, dynamic behavior analysis, and behavior surveillance. The key challenge is to effectively capture the intrinsic local-global clues of the activity. However, it remains challenging to efficiently leverage multidimensional information related to joints' local visual appearances, global spatial relationships, and coherent temporal cues. To address this challenge, we propose a joints-centered spatial-temporal feature-fused framework for action recognition, which exploits skeleton-based graph diffusion and convolution. Specifically, we employ Partial Differential Equation (PDE) based skeleton graph diffusion to automatically activate and diffuse the salient appearance features of joints. This approach simultaneously integrates the joints' appearance clues and their hierarchical relationships at both the super-pixel level and structure level. The diffused appearance-related features of the joints are further fused with skeleton-related spatial-temporal features, and the resulting fused features are fed into a skeleton convolution network for action recognition. Our method was extensively evaluated on two public datasets (NTU-RGBD and UWA3D), and the results demonstrate the improved accuracy and effectiveness of our approach. Our code will be public.

In population-oriented ergonomics product design and musculoskeletal kinetics analysis, digital spine models of different shape, pose and material property are in great demand. The purpose of this study was to construct a parameterized finite element spine model with adjustable spine shape and material property. We used statistical shape model approach to learn inter-subject shape distribution from 65 CT images of training subjects. Second order polynomial regression was used to model the age-dependent changes in vertebral material property derived from spatially aligned CT images. Finally, a parametric spine generator was developed to create finite element instances of different shapes and material properties. For quantitative analysis, the generalization ability to emulate spine shapes of different people was evaluated by fitting into 17 test CT images. The median fitting accuracy was 0.8 for Dice coefficient and 0.43 mm for average surface distance. The age-dependent bone density regression curve was also proved to well agree with large population statistics data. Finite element simulation was performed to compare how shape parameters influenced the biomechanics distribution of spine. The proposed parametric finite element whole spine model will assist the design process of new devices and biomechanical simulation towards a wide range of population.

Rapid pose classification and pose retrieval in 3D human datasets are important problems in shape analysis. In this paper, we extend the Multi-View Convolutional Neural Network (MVCNN) with ordered view feature fusion for orientation-aware 3D human pose classification and retrieval. Firstly, we combine each learned view feature in an orderly manner to form a compact representation for orientation-aware pose classification. Secondly, for pose retrieval, the Siamese network is adopted to learn descriptor vectors, where their $L_2$ distances are close for pairs of shapes with the same poses and are far away for pairs of shapes with different poses. Furthermore, we also construct a larger 3D Human Pose Recognition Dataset (HPRD) consisting of 100,000 shapes for the evaluation of pose classification and retrieval. Experiments and comparisons demonstrate that our method obtains better results than previous works of pose classification and retrieval on the 3D human datasets, such as SHREC’14, FAUST, and HPRD.