Search Results (2)

Image segmentation is a crucial task in artificial intelligence fields such as computer vision and medical imaging. While convolutional neural networks (CNNs) have achieved notable success by learning representative features from large datasets, they often lack geometric priors and global object information, limiting their accuracy in complex scenarios. Variational methods like active contours provide geometric priors and theoretical interpretability but require manual initialization and are sensitive to hyper-parameters. To overcome these challenges, we propose a novel segmentation approach, named PolarVoting, which combines the minimal path encoding rich geometric features and CNNs which can provide efficient initialization. The introduced model involves two main steps: firstly, we leverage the PolarMask model to extract multiple source points for initialization, and secondly, we construct a voting score map which implicitly contains the segmentation mask via a modified circular geometric voting (CGV) scheme. This map embeds global geometric information for finding accurate segmentation. By integrating neural network representation with geometric priors, the PolarVoting model enhances segmentation accuracy and robustness. Extensive experiments on various datasets demonstrate that the proposed PolarVoting method outperforms both PolarMask and traditional single-source CGV models. It excels in challenging imaging scenarios characterized by intensity inhomogeneity, noise, and complex backgrounds, accurately delineating object boundaries and advancing the state of image segmentation. Full article

Human poses are difficult to estimate due to the complicated body structure and the self-occlusion problem. In this paper, we introduce a marker-less system for human pose estimation by detecting and tracking key body parts, namely the head, hands, and feet. Given color and depth images captured by multiple red, green, blue, and depth (RGB-D) cameras, our system constructs a graph model with segmented regions from each camera and detects the key body parts as a set of extreme points based on accumulative geodesic distances in the graph. During the search process, local detection using a supervised learning model is utilized to match local body features. A final set of extreme points is selected with a voting scheme and tracked with physical constraints from the unified data received from the multiple cameras. During the tracking process, a Kalman filter-based method is introduced to reduce positional noises and to recover from a failure of tracking extremes. Our system shows an average of 87% accuracy against the commercial system, which outperforms the previous multi-Kinects system, and can be applied to recognize a human action or to synthesize a motion sequence from a few key poses using a small set of extremes as input data. Full article