Search

Hashing code similarity search is an effective method for large-scale image retrieval, since it provides fast search calculating and memory operating. In image retrieval, hash code based on deep convolutional networks can learn accurate image semantic information and compact hash code simultaneously. The existing methods generally extract the overall features of the image and generate corresponding hash codes. When the image contains multiple targets, this method may be suboptimal for describing the image, thus, limiting the accuracy of multi-label images. This paper focuses on the use of deep convolutional hash coding for multi-label image problems. In order to solve this problem, the authors propose a deep convolutional neural network structure suitable for multi-label image feature extraction. This network can also be used for end-to-end feature learning and further improve image retrieval accuracy. In this method, each target in the image is represented by a hash code and extensively evaluated across multiple data sets, making substantial improvement over the most advanced supervised and unsupervised hashing methods.

It is difficult to accurately segment the tongue due to the blurred contours and the similar colors of the surrounding tissue. This paper presents a semantic segmentation method for tongue body based on fully convolutional neural network. First, the convolutional and pooling layers are used to map the tongue image from the pixel domain to the feature domain. Then, the deconvolutional layer is used to restore the shrinking feature image to the original size. Finally, the Softmax classifier is trained to classify each point of the image in order to achieve segmentation. Experimental results show that the method proposed in this paper can get better segmentation results, laying a solid foundation for tongue analysis and dialectical diagnosis.

For the on-line measurement of the fabric color in the batch dyeing process, a novel method is proposed based on soft-sensor. In this method, the concentrations of the dyestuff in a jet dyeing machine are selected as the soft sensor's inputs, and the the fabric color is the output. A nonlinear model, which relates the absorbance with the concentrations of the dyestuff, is obtained based on the experiments and the regression analysis method. Using this nonlinear model, the concentrations of the dyestuff during the dyeing process are measured by the extended Kalman filter(EKF). Further, the soft-sensing model is proposed to predict the fabric color by the obtained concentrations of the dyestuff. The numerical results for the soft-sensing method of color measurement in the dyeing process of knitted cotton using reactive dyes are compared to the experimental results, leading to good agreement between theory and experiment. This demonstrates that this soft-sensing model is able to predict the fabric color in the textile industries for the batch dyeing process.

The image feature matching is a critical step in photography measurement and rebuilding. Based on the Characteristics of topography measurement, a new topography feature matching algorithm is presented in this paper. In the algorithm, the stereo images to be matched are partitioned, and then the region matching process is used to these partitioned image pairs. Based on the region matching, the corner feature matching operation is then processed, which involves relaxed iteration algorithm, zero-crossing method and least squares median algorithms to eliminate the ill matching points. Three groups of topography photography measurement images are used in image matching experiment, the experiment results shows that this algorithm can effectively improve the computation efficiency and matching accuracy, thus has a wide applications in the area based on the principle of stereo vision, such as: topography photography measurement, urban planning, etc.

In the real-time panoramic images processing, omni-directional vision sensors (ODVS) is a fast and reliable means of visual information collection, but current ODVS have a dead angle. According to the average angular resolution for two-catadioptric ODVS imaging theory, a new design of ODVS structure and catadioptric mirror is presented in this paper to overcome the defect. The reflectors curvilinear numeric solutions of this catadioptric mirror is obtained by the fourth order Runge-Kutta algorithm, and the dead angle part of the panoramic images that normal ODVS canpsilat acquire is acquired by a combine-lens of a wide-angle lens and a normal ODVS. The experimental results show that the ODVS presented in this paper can achieve a true omni-directional surveillance, and have a good applicational value in the fields such as intelligent omni-directional video surveillance, dynamic image understanding and computer vision.

In order to achieve the real-time, convenient, automatic and intelligent 3D vision measurement, this paper proposes a 3D vision measurement method based on binocular omni-directional vision sensor (ODVS), which can get the 3D panoramic image in real-time, this paper considers the human eyes as the center of vision space. At the process of gathering, processing, describing and expressing of data, this paper adopts a kind of uniform coordinate based on Gauss spherical coordinates, adopts a kind of design using ODVS having average angle resolution, and combines two ODVS having the same parameters into one. In this paper, we present the design method and experimental results. The experimental results indicate that our research results, can simplify the complexity of 3D visionpsilas measurement, can make feature extraction more convenient, can make the realization of the match of 3D images easier, and finally achieves a 3D vision measurement which is high efficient, real-time and accurate. The paper has important value on stereoscopic 3D measurement and 3D image reconstruction.

To effectively realize the image feature matching for geomorphic reverse measurement and rebuilding, a new matching scheme is presented, where the SIFT method are adopted to implement initial geomorphic image matching by going through five stages: scale-space construction, scale-space extrema detection, orientation assignment, keypoint descriptor and feature vector matching. Then, in order to eliminate the wrong matching features existing in the initial matching process, RANSAC algorithm is applied. The experimental results show that this algorithm can effectively improve the accuracy and efficiency of geomorphic image matching.

One of the most important aspects of Gerontechnology is to support the elder who lives alone at home. According to the elder's relatively obvious and stable routine of daily activities, this paper proposes a statistical approach based on machine vision to build an Elder Indoor and Outdoor Activity Model (EIOAM) through analysis of the elder's daily activity data in the main activity places and the entry/exit places collected by the Omni-directional Vision Sensor (ODVS). Since the elder's daily routine varies with the alternating of seasons and growing of age, the model should learn the routine of activities adaptively. This model is able to detect and predict the abnormal activities of the elder through detecting the significant deviations of the activity data in spatial and temporal aspects. By using this model, one can not only detect the abnormal activities happened in sight of the vision sensor (indoor), but also out of sight (outdoor). Thus provides a new methodology for the remote home care of the elder who lives alone.

Omni-directional vision sensor (ODVS) is one of important components of sensor networks. Design of ODVS, embedded Linux ODVS, video server of ODVS, and software development of ODVS applications are studied in this paper. In addition, the applications of multi-object automatic tracking systems based on ODVS are described in details. Moreover, experimental results are proposed. This research is significant for sensor networks to be efficiently and effectively implemented.

Water quality monitoring is of great importance in water safety protection and water emergency supervision. Water quality can be reflected by aquatic creatures' ecological status, especially fish's, directly and synthetically. Therefore, aquatic creatures' ecological status can be used to monitor water quality safety. This paper presents an real-time water quality monitoring technology in which computer image processing technique, network communication technique and observation of aquatic creatures' ecological status, such as small fish's, technique are used. A machine vision based biological water quality monitoring equipment is proposed. It monitors water quality safety by observing fish's biological status. This new water quality monitoring technology overcomes the shortages of existing water quality monitoring technology, which is costly, and inapplicable to large-scale long-term and on-line water quality monitoring. Thus, it provides an economical and effective method for water quality monitoring, water safety protection and water emergency supervision.

One kind of JSP automatic generation technology is introduced here, which is based on mapping mechanism. Supported by this technology, the designers who don't know JSP can participate in JSP application development. According to the mapping information setting up in advance, the system produces JSP file automatically from the HTML file. The main goal of this technology is that the ordinary Web designers who don't know programming can effectively participate in the Web application development which is based on JSP technology