Dongyue Chen

Scene classification based on gray level-gradient co-occurrence matrix in the neighborhood of interest points

Scene classification is an important application field of multimedia information technology, whereas how to extract features from image is one of the key technologies in scene classification and recognition. A new method of extracting features is presented in this paper, it extracts features through gray level-gradient co-occurrence matrix in the neighborhood of interest points, also it can reserve the key image edge information, and it is called GGNP for short in the paper. The weighted Gowers similarity coefficient model is adopted as the basis for image scene classification, as it is more flexible than Euclidean distance function. Compared with traditional methods, the method has a good invariance in image scaling, rotation, translation and robust across a substantial range of affine distortion, meanwhile having good real-time. Experimentations are designed to test the precision and time-consuming of the method, the results of experiments show that the method has good effects on scene classification.

Level Set Based Coverage Holes Detection and Holes Healing Scheme in Hybrid Sensor Network

A fundamental issue in sensor network is the coverage problem. Since the distribution of sensor nodes is not usually uniform due to random deployment and node failures, the coverage holes are hardly avoided in sensor network. And the coverage holes are important health indicators of the sensor network. This paper firstly proposes a level set based coverage holes detection algorithm for hybrid sensor network. This algorithm could estimate the number of holes and the size of the holes. Then we propose genetic algorithms based coverage holes healing algorithm. This algorithm could leverage mobility to optimize the average coverage rate and the average movement distance of the mobile nodes. Simulation results show that the proposed method could detect the holes efficiently. The holes healing algorithm outperforms the Random and Delaunay methods.

A remote sensing image matching algorithm based on the feature extraction

In this paper, a novel method for remote sensing image matching through mean-shift is proposed. First, state of the improved Mean-shift is reminded. Primary mean-shift algorithm is only based on color feature, but color feature does not apply to the remote sensing images matching. This paper exhibits a method to solve this problem using the gradient direction histogram instead of the color histogram. Secondly, Speeded-Up Robust Features (SURF) is applied to the fine matching. The experimental results show that the improved mean-shift matching algorithm, combining to the surf detector can realize two images matching accurately.

A novel method of river detection for high resolution remote sensing image based on corner feature and SVM

In this paper, a new method to detect rivers in high resolution remote sensing images based on corner feature and Support Vector Machine (SVM) is presented. It introduces corner feature into river detection for the first time. First, we detect corners in sample images and test images, and extract image corner feature with all the corners detected above. Then the corner feature and other feature of sample images, for example texture feature and entropy feature, are input into SVM for training. At last we obtain the water decision function, with which we classify each pixel into river region or background region. This method comprehensively utilizes the corner, entropy and texture feature of remote sensing images. Experimental results show that this method performances well in river automatic detection of remote sensing images.

Lung nodules classification based on growth changes and registration technology

Benign and malignant lung nodules classification is an important task in the diagnosis of lung cancer. In this study, lung nodules are classified based on growth changes feature and registration technique. Firstly, this paper combine the global rigid registration with local elastic registration method, which can extract the growth changes of a region of interest. Secondly, the benign and malignant nodules are classified on a rule-based classifier. Experimental findings show that the proposed method can extract features automatically and yield accurate classification results.

Robust object tracking with multiple basic mean shift tracker

We propose a novel tracking algorithm which can work robustly under complex dynamic scenarios. Our algorithm is based on a scheme of multiple basic mean shift tracking. In this scheme, we use Sparse Principal Component Analysis to generate multiple target models, with which each basic mean shift tracker runs in parallel at the same time. The best configuration of a target is obtained by the weighted linear combination of its basic results. In addition, for the problem that the histogram of gradient under the mean shift tracking framework is easy to fall into local maxima, we introduce the histogram of Gradient Vector Flow to represent the target. Experimental results show that our tracker is able to handle severe appearance change and recover from drifts in realistic videos. The algorithm proposed in this paper can track the target accurately and reliably compared with other existing state-of-the-art tracking algorithms.

A new method of edge detection based on PSO

Applying an edge detector to an image, in the ideal case, may obtain a set of connected curves which indicate the boundaries of objects. Actually edges in an image are a collection of pixels which are recognized as an edge in surface orientation. This paper proposes a new edge detect algorithm which uses PSO (Particle Swarm Optimization) for detection of best fitness curves in an image that represent boundaries of objects. To improve the speed of edge use the PSO on the pixels whose gradient grate than the threshold. Use image with simple geometric objects, with impulse noise levels and the image have complex texture to assess the system. Use this algorithm on the images with high noise levels to detect edge is more accurately than existing edge detector.

Fast scene recognition based on saliency region and SURF

Scene recognition is a hot topic in the field of computer vision, a fast scene recognition method based on saliency region and SURF (speeded up robust features) is proposed in this paper. This method adopts PFT (phase fourier transform) to construct saliency map, on the basis the algorithm of top-ranking extreme points selection based neighborhood entropy is used get saliency region information. Finally scene recognition is implemented using SURF of the saliency region. The method effectively improves real-time of scene recognition and the capability of scene analysis. Compared with other scene recognition methods, it has a better invariance in image rotation, scaling, translation and a substantial range of affine distortion, meanwhile having better real-time. The results of experiments with university of Southern California scene database demonstrate that the method performed well in recognition result, computational speed and robustness.

Distance-based over-segmentation for single-frame RGB-D images

Over-segmentation, known as super-pixels, is a widely used preprocessing step in segmentation algorithms. Oversegmentation algorithm segments an image into regions of perceptually similar pixels, but performs badly based on only color image in the indoor environments. Fortunately, RGB-D images can improve the performances on the images of indoor scene. In order to segment RGB-D images into super-pixels effectively, we propose a novel algorithm, DBOS (Distance-Based Over-Segmentation), which realizes full coverage of super-pixels on the image. DBOS fills the holes in depth images to fully utilize the depth information, and applies SLIC-like frameworks for fast running. Additionally, depth features such as plane projection distance are extracted to compute distance which is the core of SLIC-like frameworks. Experiments on RGB-D images of NYU Depth V2 dataset demonstrate that DBOS outperforms state-ofthe-art methods in quality while maintaining speeds comparable to them.

A new face mesh model based on edge attractor and nonlinear global topological constraints

Face Feature Point Detection (FFPD) is a significant and interesting topic in many related areas of face recognition. A new face mesh model is presented in this paper to realize the FFPD in a fast and training-free approach. At first, an automatic face mesh initialization algorithm based on the detection of face, eye-pairs and mouth is proposed. Second, a local operator referred as Edge Attractor and an adaptation algorithm based on global topological constraint are proposed to locate feature points at boundaries of facial features while maintain the global shape of the face mesh. At last, an iterative algorithm is developed to optimize the evaluation function of the face mesh. Experimental results showed that the proposed face mesh model is fast, robust, and accurate in detecting face feature points.