In the oil and gas related fields, three-phase distribution flow in pipelines is one of the important aspects that need to be continuously monitored in the Industrial Pipeline Internet of Things (IoT). The dual-modality system of Electrical Capacitance Tomography (ECT) and Electromagnetic Tomography (EMT) can reconstruct the three-phase distribution within a pipeline’s cross-section. However, due to mutual interference among three-phase flow media and noise effects, the reconstruction quality of ECT and EMT images is degraded. In this paper, an ECT-EMT image fusion method based on cross-sensitive field optimization and super-resolution preprocessing is proposed. Firstly, unimodal ECT image reconstruction generates virtual image in solid-phase fluid regions, leveraging this information, a cross-sensitive field optimization method combined with the YOLOv8 network is proposed to improve the EMT reconstructed image quality. Compared with traditional fixed-sensitivity-field EMT reconstruction methods, the correlation coefficient increased by an average of 3.8%. Then, a Gaussian second-order derivative super-resolution method is introduced, which uses a Gaussian second-order derivative function to fit fuzzy kernel parameters and deconvolves the degradation matrices of ECT and EMT images, solving the edge-blurring problem in their fusion. Compared to traditional deblurring methods such as Wiener filtering and single Gaussian kernel modelling, PSNR improved by an average of 1.4 dB, while SSIM improved by an average of 0.03. Finally, an image fusion method based on Latent Low-Rank Representation is proposed by extracting global and local feature information from super-resolution preprocessed ECT and EMT grayscale matrices, respectively. Compared to mainstream pixel-level fusion methods such as wavelet transforms, it resolves the issue of performance degradation in traditional fusion methods when observational data is insufficient. All validations were conducted using simulation data from COMSOL, with sensitivity field optimisation adapted to the complex flow patterns of three-phase flow scenarios. The super-resolution module addressed reconstructed images exhibiting low signal-to-noise ratios and blurred edges, whilst the LatLRR fusion method maintained stable fusion performance even under small-sample conditions.

The volume of Web videos have increased sharply through the past several years because of the evolvement of Web video sites.Enhanced algorithms on retrieval, classification and TDT (abbreviation of Topic Detection and Tracking) can bring lots of convenience to Web users as well as release tedious work from the administrators. Nevertheless, due to the the insufficiency of annotation keywords and the gap between video features and semantic concepts, it is still far away from satisfactory to implement them based on initial keywords and visual features. In this paper we utilize a keyword propagation algorithm based on manifold structure to enrich the keyword information and remove the noise for videos. Both text similarity and temporal similarity are employed to explore the relationship between any pair of videos and to construct the propagation model. We explore three applications, i.e., TDT, Retrieval and Classification based on a Web news video dataset obtained from a famous online video-distributing website, YouKu, and evaluate our approach. Experimental results demonstrate that they achieve satisfactory performance and always outperform the baseline methods.

Name ambiguity stems from the fact that many people or objects share identical names. In this paper, we focus on investigating the problem in digital libraries to distinguish publications written by authors with identical names. We present an effective graph-based framework, GHOST (abbr. GrapH-based framewOrk for name diStincTion), to solve the problem systematically. We evaluated the framework on the real DBLP dataset, and the experimental results show that GHOST outperforms the state-of-the-art method.

The new media initiative project of Ryerson University had been successful in applying image and video processing techniques in creating special effects in film making. Our system utilizes a graph cut image segmentation and snakes active contour approach to obtain object cut outs in 3-D with tracking. It uses exemplar-based inpainting for background filling for missing regions uncovered by object transformation. The two methods combined allows for easy video object editing with reduction in user input. Previously implemented shot detection by twin window amplification method and steerable pyramids texture generation is also part of the system. Lastly, a set of transformation was implemented that serves as a basis for testing the system.

Visual-based wireless sensor networks have been implemented in several different fields such as environment monitoring, military applications, and robotic applications. Due to the limitation of node's specification, the bandwidth and energy become critical issues for sensor nodes. In this paper, we employ a solar cell recharging model and a layered clustering model to deal with the restrict energy consumption under the consideration of visual quality. The system lifetime can be prolonged by rechargeable solar cell that can be recharged by solar panel in daytime. In addition, we analyze the simulation results of energy consumption and total transmitted packets by changing the aggregation rate and gate energy (GE). With the aggregation rate decreasing, the cluster head in inner layer can support more visual nodes and reserve more bandwidth. The lower GE can reduce the packets loss during the system charging process. The analysis and experiment result obtained in this paper prove that with the combination of layered clustering and solar recharging, the performance of wireless visual sensor network can be enhanced under the consideration of the restrict node's capacity and video distortion.

At present, the researchers are making great effort to search for a general purpose resolution for image segmentation. Most of the approaches we are using now are somewhat inefficient and tardy. In this paper we present an innovative method, parallel self-organizing tree map (PSOTM), which is improved from SOTM. By processing image in parallel, PSOTM could segment the image in a much faster processing speed. After applying the new technology, PSOTM, we could obtain the higher efficiency with limited impacts on the visual quality