Final Matching Research Articles

INDOOR SCENE REGISTRATION BASED ON SIAMESE NETWORK AND POINTNET

Abstract. This paper presents a deep learning feature-based method for registration of indoor mobile LiDAR data. Our method is to input point cloud directly, which is more robust to noise than traditional algorithms. The proposed method involves three steps. We first extract the key points by Harris3D algorithm and get their local patches by our sampling method. Second, a Siamese network is trained to describe the patches as local descriptors. Finally, we obtain the final matching pairs depends on the distance which is between two descriptors, and then solve the transformation matrix. The accuracy of registration is within 6 cm when the overlap is greater than 35%. In order to improve the registration accuracy, the ICP algorithm is used to fine-tuning the registration results. And the final registration accuracy is within 3.5 cm. The experiments show that our method applied to the registration of indoor mobile LiDAR data robustly and accurately.

Intuitionistic fuzzy two-sided matching model and its application to personnel-position matching problems

Personnel-position matching (PPM) problems have gained wide attention from scholars in the past decades. However, the traditional matching methodologies cannot be used to solve the two-sided matching problem between personnel and position effectively. In this article, we design an intuitionistic fuzzy two-sided matching model for the PPM problem. In order to describe the correlations between the evaluated attributes, some novel intuitionistic fuzzy Choquet integral aggregation operators are proposed. The final optimal matching results can be obtained by solving the multiobjective decision-making model. A case about the PPM problem is presented to illustrate the practicality of this method. The results of this article indicate that the intuitionistic fuzzy Choquet integral aggregation operators-based two-sided matching model can effectively solve the PPM problems where their evaluated attributes are correlated.

Time domain predictions of inertial loads on a drifting ship in irregular beam waves

A proper estimation of the ship accelerations, particularly in the transversal direction, is fundamental for predicting the risk of cargo shifts and inherent structural damages in severe irregular seas. A specific situation worth to be considered in this context is the so-called dead ship condition, in which the ship, because of a loss of steering capabilities, is drifting in beam waves at zero speed. Particularly in this situation, a proper modelling of ship dynamics in wave is deemed necessary for a reliable assessment of inertial loads.In this paper three time domain procedures with different accuracy are applied and compared. The aim is to investigate the dependency of inertial loads on the non-linear effects of ship geometry and on the couplings among the 6DOF of the ship.Simulations are carried out with, in input, the same irregular sea realizations in the time domain, obtained from a digitalisation of JONSWAP spectra. A technique allowing for a fine matching of the digitalised spectra to the analytical ones is developed and applied.The investigation indicates that couplings among ship motions trigger larger inertial loads and, accordingly, a larger probability of occurrence for cargo shifting and structural damages.

Fast large scale deep face search

Towards the large scale face search problem, this paper proposes a fast deep face search method which is realized by combination of deep convolution neural network (CNN), semantic hashing, and hash-based similarity search. First of all, to boost the performance in accuracy of face search, the residual network (Resnet) is exploited to construct a deep face feature model and then train it over the cleaned MS-Celeb-1M, which is used to extract real-valued face feature. Next, by imposing PCA and binarization operations, we convert the real-valued feature into compact hash code used for speeding up the face search. Based on the extracted dual features, the face search can be efficiently performed by adopting two-stage matching (i.e., coarse matching and fine matching) strategy. The coarse matching is implemented under the support of efficient hash indexing technique for yielding a small number of candidates while the fine stage is to filter out the unrelated images by cosine distance comparison of real-valued features. It is worth noting that we offer two coarse matching methods, such as GPU-hash and M-index-hash based matching, which are suitable for tens-of million and billion scale scenarios respectively. The experimental results demonstrate that the proposed method is very effective for large scale face search in both aspects of accuracy and real time property.

Bi-directional attention comparison for semantic sentence matching

Semantic sentence matching, also known as calculation of text similarity, is one of the most important problems in natural language processing. Existing deep models mostly focus on the neural networks with attention mechanism. In this paper, we present a deep architecture to match two Chinese sentences, which only relies on alignment instead of long short-term memory network after attention mechanism is employed to get interaction information between sentence-pairs, the model becomes more lightweight and simple. Meanwhile, in order to capture semantic features enough, in addition to using max pooling and average pooling operation, we also employ a pooling operation named attention-pooling to aggregate information from the whole sentence, the final matching score is obtained after a multilayer perceptron classifier. Experiments are carried out on ATEC-NLP dataset and outline the effectiveness of our approach.

Sequential Far Infrared Image Mosaic Using Coarse-to-Fine Scheme

Sequential far infrared (FIR) image mosaic is a challenging task due to repeated patterns, low contrast, image blurring, illumination change, and sparse structure. In this paper, a novel matching approach is proposed for the sequential FIR image registration, which is based on coarse-to-fine strategy and grid-based motion statistics (GMS). In the coarse matching step, a transitive relation is deduced to obtain the coordinate relations of the current registering image and its adjacent images to ensure enough number of feature points and guarantee the accuracy of the preregistration results. In the fine matching step, a principle for fine-tuning registration and a strategy for changing the distribution of the feature points are developed for finding most appropriate feature points. Based on the point pairs, the position of the current registering image can be accurately located. Besides, an effective post-processing, such as gray-scale non-uniformity correction and mosaic seam elimination, is also utilized to refine the fine-tuning results to achieve the promising mosaic effect. A comparative study and quantitative evaluation with commercial software or representative image mosaic algorithm demonstrate that better quality results can be obtained by using the proposed method.

Multi-Turn Response Selection for Chatbots With Hierarchical Aggregation Network of Multi-Representation

Matching an appropriate response with its multi-turn context is a crucial challenge in retrieval-based chatbots. Current studies construct multiple representations of context and response to facilitate response selection, but they use these representations in isolation and ignore the relationships among representations. To address these problems, we propose a hierarchical aggregation network of multi-representation (HAMR) to leverage abundant representations sufficiently and enhance valuable information. First, we employ bidirectional recurrent neural networks (BiRNN) to extract syntactic and semantic representations of sentences and use a self-aggregation mechanism to combine these representations. Second, we design a matching aggregation mechanism for fusing different matching information between each utterance in context and response, which is generated by an attention mechanism. By considering the candidate response as the real part of the context, we try to integrate all of them in chronological order and then accumulate the vectors to calculate the final matching degree. An extensive empirical study on two multi-turn response selection data sets indicates that our proposed model achieves a new state-of-the-art result.

A Study on Detection and Recognition of Facial Area Using Linear Discriminant Analysis

We propose a more stable robust recognition algorithm which detects faces reliably even in cases where there are changes in lighting and angle of view, as well it satisfies efficiency in calculation and detection performance. We propose detects the face area alone after normalization through pre-processing and obtains a feature vector using (PCA). The feature vector is applied to LDA and using Euclidean distance of intra-class variance and inter class variance in the 2nd dimension, the final analysis and matching is performed. Experimental results show that the proposed method has a wider distribution when the input image is rotated 45 ° left / right. We can improve the recognition rate by applying this feature value to a single algorithm and complex algorithm, and it is possible to recognize in real time because it does not require much calculation amount due to dimensional reduction.

Model updating of a historic concrete bridge by sensitivity- and global optimization-based Latin Hypercube Sampling

In this paper, a self-implemented model updating global optimization procedure is successfully applied to a remarkable case study concerning a historic centennial Reinforced Concrete (RC) bridge with parabolic arches, based on recorded experimental vibrational data and arising identification of modal properties. In order to boost the degree of confidence and robustness of the developed model updating procedure, appropriate computational strategies are proposed at the level of both Sensitivity Analysis (SA) and global optimization. In particular, Latin Hypercube Sampling (LHS) is employed in drawing up both strategies, as a systematic automated way to determine appropriate multi-start sets of initiation points, optimally distributed throughout the parametric domain. The procedure involves a gradient-based method and proposes an interaction algorithm between mechanical FEM solver and numerical computing environment. Moreover, the gradient of the objective function involved in the model updating is analytically derived, instead of by often-used Finite Differences (FD), toward better accuracy and computational efficiency. Comprehensive updating results starting from a first FEM base model are achieved, for the considered case study, and show that the relative eigenfrequency and mode shape estimations are considerably improved, for all the structural modes accounted for within the updating process, with a very good final matching between experimentally extracted and FEM modelled modal properties.

Data-Driven Optimization Control for Dynamic Reconfiguration of Distribution Network

To improve the reliability and reduce power loss of distribution network, the dynamic reconfiguration is widely used. It is employed to find an optimal topology for each time interval while satisfying all the physical constraints. Dynamic reconfiguration is a non-deterministic polynomial problem, which is difficult to find the optimal control strategy in a short time. The conventional methods solved complex model of dynamic reconfiguration in different ways, but only local optimal solutions can be found. In this paper, a data-driven optimization control for dynamic reconfiguration of distribution network is proposed. Through two stages that include rough matching and fine matching, the historical cases which are similar to current case are chosen as candidate cases. The optimal control strategy suitable for the current case is selected according to dynamic time warping (DTW) distances which evaluate the similarity between the candidate cases and the current case. The advantage of the proposed approach is that it does not need to solve complex model of dynamic reconfiguration, and only uses historical data to obtain the optimal control strategy for the current case. The cases study shows that the optimization results and the computation time of the proposed approach are superior to conventional methods.

Multi-sensor SAR image coregistration based on stable convex corner points and double matching network

ABSTRACTSince significant differences exist in imaging principle and parameters of different synthetic aperture radar (SAR) sensors, the detection of stable consistent features and accurate feature matching are difficult to implement in multi-sensor SAR image coregistration. To solve these problems, this study proposed a coregistration method based on stable convex corner point and double matching network. The stable convex corner points were detected from stable convex polygons, which were obtained by the proposed local stable extremal region extraction method, convex polygon fitting, and normalization. The double matching network was divided into coarse and fine matching. We utilized spatial relationships of the stable convex polygons to build a coarse matching network, and the fine matching network was formed by stable convex corner points. Precise matching was achieved with region features and spatial relations. Experimental results on multi-sensor SAR images showed that the proposed method had superior performance.

A user identity matching method based on integrating account attributes

Aiming at the low utilization rate of attribute information and the lack of mining of the correlation among attributes of the existing cross-social network user identity matching algorithms, we proposed an algorithm for user identity matching across social networks utilizing fuzzy measure and Choquet integrals. Firstly, according to the characteristics of different attributes, we determined different similarity calculation strategies; Secondly, we utilized particle swarm optimization method to calculate the fuzzy density of each attribute; Then Choquet integral was utilized to calculate the similarity of two accounts; Finally, the similarity was compared with the preset matching threshold and the final matching result was obtained. The experimental results in multiple sets of data showed that the average F1 value of the proposed algorithm reaching 84.5%. The performance is not only better than traditional machine learning methods, but also better than several baseline algorithms. It can be more accurate to identify the same user’s accounts in multiple social networks according to the attribute information.

Ear recognition in 3D using 2D curvilinear features

This study presents a novel approach for human recognition using co-registered three-dimensional (3D) and 2D ear images. The proposed technique is based on local feature detection and description. The authors detect feature key-points in 2D ear images utilising curvilinear structure and map them to the 3D ear images. Considering a neighbourhood around each mapped key-point in 3D, a feature descriptor vector is computed. To match a probe 3D ear image with a gallery 3D ear image for recognition, first highly similar feature key-points of these images are used as correspondence points for an initial alignment. Afterwards, a fine iterative closest point matching is performed on entire data of the 3D ear images being matched. An extensive experimental analysis is performed to demonstrate the recognition performance of the proposed approach in the presence of noise and occlusions, and compared with the available state-of-the-art 3D ear recognition techniques. The recognition rate of the proposed technique is found to be 98.69% on the University of Notre Dame-Collection J2 dataset with an equal error rate of 1.53%.

Synthetic aperture radar image registration based on hybrid feature detection and multifeature constraint matching

The alignment of images acquired using the synthetic aperture radar (SAR) is a central task in image processing for remote sensing. Owing to speckle noise and the particular features of SAR imaging, stable feature detection and accurate matching remain challenging tasks. We propose a registration method for SAR images. In feature detection, we develop a hybrid feature detection method to identify structural and textural features. A stable convex corner point feature is detected using stable convex polygons obtained by a proposed method for local stable extremal region extraction and convex polygon fitting. The stable local extremum points in each convex polygon are detected by an improved scale invariant feature transform method. In feature matching, a multifeature constraint matching method is designed for accurate matching. Coarse matching is implemented using the constraints on the region and its shape as well as the network. The use of stable local extremum points with spatial constraints can eliminate mismatches and yield a fine match. The results of experiments verified the effectiveness and accuracy of the proposed method.

Posture adjustment of workpiece based on stepwise matching by self-adaptive differential evolution algorithm

Abstract. The workpiece contour errors from previous process affect current and subsequent process. In order to improve the uniformity of workpiece allowance distribution, a stepwise workpiece matching adjustment method with contact inspection is developed. This method includes two stepwise registration processes. At first, some pairs of measured points on the theoretical and actual surfaces are selected to build the corresponding local coordinate systems, and then the rough matching matrixes are obtained by the coordinate systems alignment. During the fine matching process, the objective function based on the least square method is established by the measured point sets adjusted through the rough matching. The fine matching matrixes can be obtained by self-adaptive differential evolution algorithm. The posture adjustment can be realized by transforming coordinate systems of 5 axis machine tool, of which adjusted values can be calculated by the matching matrixes and the machine tool topology. At last, some experiments were presented to demonstrate the performance of the method.

Multiple domains of bacterial and human Lon proteases define substrate selectivity

The Lon protease selectively degrades abnormal proteins or certain normal proteins in response to environmental and cellular conditions in many prokaryotic and eukaryotic organisms. However, the mechanism(s) behind the substrate selection of normal proteins remains largely unknown. In this study, we identified 10 new substrates of F. tularensis Lon from a total of 21 candidate substrates identified in our previous work, the largest number of novel Lon substrates from a single study. Cross-species degradation of these and other known Lon substrates revealed that human Lon is unable to degrade many bacterial Lon substrates, suggestive of a “organism-adapted” substrate selection mechanism for the natural Lon variants. However, individually replacing the N, A, and P domains of human Lon with the counterparts of bacterial Lon did not enable the human protease to degrade the same bacterial Lon substrates. This result showed that the “organism-adapted” substrate selection depends on multiple domains of the Lon proteases. Further in vitro proteolysis and mass spectrometry analysis revealed a similar substrate cleavage pattern between the bacterial and human Lon variants, which was exemplified by predominant representation of leucine, alanine, and other hydrophobic amino acids at the P(−1) site within the substrates. These observations suggest that the Lon proteases select their substrates at least in part by fine structural matching with the proteins in the same organisms.

Composite components-based face sketch recognition

Abstract Composite face sketch recognition is a challenging problem in face recognition, which is important for law enforcement. Considering composite sketch is generated from facial components, a component-based representation approach (CBR) was proposed to match composite sketch to photos recently. They extract multi-scale local binary pattern (MLBP) feature from every component and use the fusion of four most discriminative components as the final matching score, ignoring the inherent structure of composite sketch and the fusion does not take the full components’ information into consideration. This paper presents a novel composite sketch recognition method by extracting scale-invariant feature transform (SIFT) feature and histogram of oriented gradient (HOG) feature from components, fusing different features at score level, combining facial components with linear function. In our proposed method, feature fusion could extract local texture and structure feature from composite sketches. The linear combination not only reserves complete component information but also protrudes different components contribution to the holistic face image. In addition, we show the detailed procedure of composite sketch generation experiment, helping build related face database. Experiments on several public composite sketch databases (including a newly published UoM-SGFS database) demonstrate that our proposed method achieves superior performance compared with state-of-the-art methods.

A Coarse-to-Fine Geometric Scale-Invariant Feature Transform for Large Size High Resolution Satellite Image Registration.

Large size high resolution (HR) satellite image matching is a challenging task due to local distortion, repetitive structures, intensity changes and low efficiency. In this paper, a novel matching approach is proposed for the large size HR satellite image registration, which is based on coarse-to-fine strategy and geometric scale-invariant feature transform (SIFT). In the coarse matching step, a robust matching method scale restrict (SR) SIFT is implemented at low resolution level. The matching results provide geometric constraints which are then used to guide block division and geometric SIFT in the fine matching step. The block matching method can overcome the memory problem. In geometric SIFT, with area constraints, it is beneficial for validating the candidate matches and decreasing searching complexity. To further improve the matching efficiency, the proposed matching method is parallelized using OpenMP. Finally, the sensing image is rectified to the coordinate of reference image via Triangulated Irregular Network (TIN) transformation. Experiments are designed to test the performance of the proposed matching method. The experimental results show that the proposed method can decrease the matching time and increase the number of matching points while maintaining high registration accuracy.

Variation in Medicare Payments and Reimbursement Rates for Hepatopancreatic Surgery Based on Quality: Is There a Financial Incentive for High-Quality Hospitals?

To better define the financial impact of high-quality care for payers and hospitals, we compared outcomes and Medicare payments between high-quality (HQ) and low-quality (LQ) hospitals after hepatopancreatic surgery. Between 2013 through 2015, a total of 15,874 Medicare beneficiaries underwent hepatopancreatic surgery. Using the entire cohort, multivariable logistic regression was performed to categorize hospitals into quintiles based on the probability of experiencing a major complication; HQ (bottom 20%) and LQ (top 20%) hospitals were identified. Only HQ and LQ hospitals were included in the final propensity matching to compare payments. Major complication was defined as a complication associated with a length of stay of >75th percentile. Incremental payment and cost of complication were estimated using multivariable linear regression. Major complications occurred in 9.7% (n= 309 of 3,182) at HQ hospitals compared with 20% (n= 625 of 3,130) at LQ hospitals (p < 0.001). The incremental increased payment associated with major complication was $29,640, which was lower than the incremental hospital cost of $42,935. The Medicare reimbursement rate was also 6% lower at both HQ and LQ hospitals when a major complication occurred vs not; however, HQ hospitals had a 3% higher reimbursement rate compared with LQ hospitals when a major complication did not occur (p= 0.002). Mean unadjusted Medicare payment was lower at HQ hospitals by $5,165 per patient vs LQ hospitals (p < 0.001), largely because HQ hospitals had a lower overall incidence of major complications (n= 315 vs n= 625). By having 310 fewer patients with a major complication, HQ hospitals collectively achieved $3.1 million/year in Medicare savings. High-quality hospitals are able to achieve substantial Medicare savings by avoiding major complications. Occurrence of major complications was associated with lower Medicare reimbursement rates at both HQ and LQ hospitals vs when no complications occurred.

Effect of pore geometry on nitrogen sorption isotherms interpretation: A pore network modeling study

Hydrocarbon-bearing tight rocks usually have pores at nanometer scale and permeability below microdarcy range. An effective characterization of the pore structure is essential for understanding hydrocarbon storage and transport behaviors. Nitrogen sorption isotherms measured at 77 K have been widely used for pore size and surface area characterization of nanoporous media. However, most analysis efforts are based on the assumption of slit or cylindrical pore shapes. Unconventional rocks such as shale have complex pore geometries that can have a significant effect on the adsorption curve cumulative behavior. Thus, a proper understanding of the effect of the assumed pore shapes on the interpretation of pore structure from nitrogen sorption isotherms is required to assess the reliability of the analysis results. In this study, we apply the lattice density functional theory initially developed for a single pore to a network of pores with bimodal pore size distribution and three types of pore shapes (i.e. slit, cylindrical and spherical). We simulate the nitrogen adsorption and desorption processes with incorporation of pore blocking effect and tensile strength effect. We optimize the pore size distribution, connectivity, surface area and surface energy for each type of pore shape by matching simulation and experimental sorption curves. We apply the above workflow to the sorption curves collected on Cameo coal and Woodford shale samples. For both samples, a network of spherical pores always results in the largest average pore size, highest surface energy, highest degree of connectivity but smallest surface area, while a network of slit pores shows the opposite. At the same time, all three types of pore shapes demonstrate a good match with experimental sorption curves. In terms of providing the closest match of surface area with BET results, we determine the dominant pore shape of Cameo coal and Woodford shale samples to be slit and cylindrical, respectively. To reduce the uncertainty in the characterization results, we further use a combination of all three types of pore shapes in our pore network model, with fractions of each pore shape determined by SEM image analysis. The final matching results show significant improvement over using only slit pore shape.