Guangrong Ji

An estimation-based approach for range image segmentation: on the reliability of primitive extraction

This paper presents a new algorithm for estimation-based range image segmentation. Aiming at surface-primitive extraction from range data, we focus on the reliability of the primitive representation in the process of region estimation. We introduce an optimal description of surface primitives, by which the uncertainty of a region estimate is explicitly represented with a covariance matrix. Then the reliability of an estimate is interpreted in terms of “measure of uncertainty”. The segmentation approach follows the region-growing scheme, in which the regions are estimated in an iterative way. With the probabilistic model proposed in this paper, surface homogeneity is defined and tested by an optimal criterion. A notable feature of the algorithm is that the order of merging is organized to lead the growth towards the most reliable representation of the merged region. Concerned with man-made objects in the scene, we restrict the class of surface primitives to be quadric or planar. The proposed algorithm has been applied to real data and synthetic data and demonstrated with experimental results.

A new method for fast computation of moments based on 8-neighbor chain code applied to 2-D object recognition

2D moment invariants have been successfully applied in pattern recognition tasks. The main difficulty of using moment invariants is the computational burden. To improve the algorithm of moments computation through an iterative method, an approach for fast computation of moments based on the 8-neighbor chain code is proposed in this paper. Then artificial neural networks are applied for 2D shape recognition with moment invariants. Compared with the method of polygonal approximation, this approach shows higher accuracy in shape representation and faster recognition speed in experiments.

Subsurface object detection using UWB Ground Penetrating Radar

Ultra Wideband (UWB) Ground Penetrating Radar (GPR) has been widely used in the detection of buried Unexploded Ordnance (UXO), particularly for small and/or shallow objects. Detecting subsurface objects from a weak reflection signal against strong clutter from GPR data is an important problem. A detection method which employs Principal Component Analysis (PCA) and digital image processing techniques is proposed in this paper for clutter reduction in underground object location. The proposed method can robustly indicate not only the regular hyperbolas but also the perturbed slant lines as a result of buried objects. Experimental data collected using an UWB GPR system is used to demonstrate that the proposed methods are effective in reducing clutter and detecting subsurface targets.

Application of Connected Morphological Operators to Image Smoothing and Edge Detection of Algae

Connected morphological operators have the virtue of simplifying image while preserving the edge information. Connected morphological operators are used in combination to smooth original algae images which aims to suppress noise and to simplify image. First, area opening is applied to suppress noise and keep the algal body and thin branches which are connected with body. Second, morphological reconstruction from marker and attribute thinning are performed to extract some lost fragments of body contour and acerose spinule. The smoothed algae images preserve the majority of outline information. Thus image smoothing of algae is accompanied with detection of edges simultaneously. Experimental results demonstrate that the composite method proposed in this paper is efficient in noise elimination and edges extracted correspond well with the outlines of algae visually.

3D object recognition from 2D invariant view sequence under translation, rotation and scale by means of ANN ensemble

In this paper, we present a supervised multiple-weight neural network ensemble strategy for 3D object recognition from 2D multiple-view invariant sequence, so as to achieve omnidirectional information accumulation or solution in large-scale database. View information with transition in explicitly temporal order, is empirically selected for training set. On condition that requirements could not be met to a certain extent in one 3D object, more complicated training set is adopted in order to regrow and expand knowledge until satisfactory, without affecting knowledge acquired previously in other 3D objects. Simulation experiment for 3D object recognition from 2D view sequence achieved encouraging results, and proved effective and feasible in the approach proposed.

ZooplanktoNet: Deep convolutional network for zooplankton classification

Zooplankton are quite significant to the ocean ecosystem for stabilizing balance of the ecosystem and keeping the earth running normally. Considering the significance of zooplantkon, research about zooplankton has caught more and more attentions. And zooplankton recognition has shown great potential for science studies and mearsuring applications. However, manual recognition on zooplankton is labour-intensive and time-consuming, and requires professional knowledge and experiences, which can not scale to large-scale studies. Deep learning approach has achieved remarkable performance in a number of object recognition benchmarks, often achieveing the current best performance on detection or classification tasks and the method demonstrates very promising and plausible results in many applications. In this paper, we explore a deep learning architecture: ZooplanktoNet to classify zoolankton automatically and effectively. The deep network is characterized by capturing more general and representative features than previous predefined feature extraction algorithms in challenging classification. Also, we incorporate some data augmentation to aim at reducing the overfitting for lacking of zooplankton images. And we decide the zooplankton class according to the highest score in the final predictions of ZooplanktoNet. Experimental results demonstrate that ZooplanktoNet can solve the problem effectively with accuracy of 93.7% in zooplankton classification.

Automatic setae segmentation from Chaetoceros microscopic images

A novel image processing model Grayscale Surface Direction Angle Model (GSDAM) is presented and the algorithm based on GSDAM is developed to segment setae from Chaetoceros microscopic images. The proposed model combines the setae characteristics of the microscopic images with the spatial analysis of image grayscale surface to detect and segment the direction thin and long setae from the low contrast background as well as noise which may make the commonly used segmentation methods invalid. The experimental results show that our algorithm based on GSDAM outperforms the boundary‐based and region‐based segmentation methods Canny edge detector, iterative threshold selection, Otsus thresholding, minimum error thresholding, K‐means clustering, and marker‐controlled watershed on the setae segmentation more accurately and completely. Microsc. Res. Tech. 77:684–690, 2014.

Image Segmentation Based on Chaos Immune Clone Selection Algorithm

Image segmentation is a fundamental step in image processing. Otsus threshold method is a widely used method for image segmentation. In this paper, a novel image segmentation method based on chaos immune clone selection algorithm (CICSA) and Otuss threshold method is presented. By introducing the chaos optimization algorithm into the parallel and distributed search mechanism of immune clone selection algorithm, CICSA takes advantage of global and local search ability. The experimental results demonstrate that the performance of CICSA on application of image segmentation has the characteristic of stability and efficiency.

The prediction of the financial time series based on correlation dimension

In this paper we firstly analysis the chaotic characters of three sets of the financial time series (Hang Sheng Index (HIS), Shanghai Stock Index and US gold price) based on the phase space reconstruction. But when we adopt the feedforward neural networks to predict those time series, we found this method run short of a criterion in selecting the training set, so we present a new method: using correlation dimension (CD) as the criterion. By the experiments, the method is proved effective.

An approach of image decomposition for underwater target detection by inhomogeneous illumination based on G-Space and PDE

This paper presents a new approach of image decomposition for underwater target detection by inhomogeneous illumination based on G-Space and Partial Differential Equation( PDE). Underwater target images with high contrast visibility (less back-scattering) can be obtained within the inhomogeneous illumination field which power density is allocated inversely propotional to the rule of the light attenuation in water medium. Then the image f is decomposed into a sum of two functions u + v, where u component is modeled by a function of bounded variation (a cartoon or sketchy approximation of f), while v component representing the texture or noise is modeled by an oscillatory function. In this paper, the Vese-Osher(VO) model and Mumford-Shah-G(MS-G) model based on G-Space and PDE are introduced for texture image extraction. And the experimental results show that its effective to obtain the cartoon and texture components as well as edge component of the underwater targets by MS-G model, which can be applied to further procedures such as image reconstruction and object recognition for underwater target detection.