Search

Traditionally, multi-label feature extraction algorithm is used to extract local features of blurred images under complex illumination conditions. The computational complexity is extremely high and the extraction efficiency is low. Therefore, in this paper, the local feature intelligent extraction algorithm for blurred images under complex illumination conditions is proposed. The improved image segmentation algorithm based on local fuzzy C-means clustering is used to cluster and segment the blurred images. The image feature recognition algorithm based on wavelet transform and LBP log domain feature extraction is adopted. The blurred image is transformed from the spatial domain to the logarithmic domain to make two-stage discrete wavelet decomposition, and the high-frequency component is used to reconstruct the original image to perform high-pass filtering on the blurred image. By filtering the low-frequency illumination component to compensate the complex illumination, the block LBP is used to extract the local texture features of the blurred image after illumination compensation. The experimental results on Yale-B face database show that the proposed algorithm can effectively extract the local features of blurred images under complex illumination conditions, and the maximum robustness of the algorithm is 0.44, the maximum value of feature extraction error rate is 0.25, and the maximum value of feature extraction speed growth rate is 96%, with high robustness, accuracy and extraction efficiency.

Automatic Feature Recognition (AFR) is considered as the key connection technique of the integration of Computer Aided Design (CAD) and Computer Aided Process Planning (CAPP). At present, there is a lack of a systematic method to identify and evaluate the local features of 3D CAD models. The process information such as topological structure, shape and size, tolerance and surface roughness should be considered. Therefore, a novel Model Based Definition (MBD) based on 3D CAD model AFR and similarity evaluation are proposed in this paper. A Multi-Dimensional Attributed Adjacency Matrix (MDAAM) based on MBD is established based on the fully consideration of the topological structure, shape and size, surface roughness, tolerance and other process information of the B-rep model. Based on the MDAAM, a two-stage model local feature similarity evaluation method is proposed, which combines the methods of optimal matching and adjacency judgment. First, the faces of source feature and target model are used as independent sets to construct a bipartite graph. Secondly, supplement the vertices in the independent set of source feature to make the number of vertices in two independent sets equal. Thirdly, based on MDAAM data, the weighted complete bipartite graph is constructed with the face similarity between two independent sets as the weight. Fourthly, Kuhn-Munkres algorithm is used to calculate the optimal matching between the faces of source feature and target model. Fifthly, the adjacency between matching faces in target model is judged. Finally, the similarity between matching faces of the two models is calculated, which is used as the similarity evaluation result. The effectiveness of this method is verified by three applications.

In recent times, with the advancement of digital imaging, automatic facial recognition has been intensively studied for adults, while less for neonates. Due to the miniature facial structure and facial attributes, newborn facial recognition remains a challenging area. In this paper, an automatic video-based Neonatal Face Attributes Recognition (NFAR) approach in a hierarchical framework is proposed by coalescing the intensity-based method, pose estimation, and novel dedicated neonatal Face Feature Selection (FFS) algorithm. The intensity-based method is used for face detection, followed by the facial pose estimation algorithm and FFS are dedicated to neonatal pose and face feature recognition, respectively. In this study, video-data of 19 neonates' were collected from the Children's Hospital affiliated to Fudan University, Shanghai, to evaluate the proposed NFAR approach. The results show promising performance to detect the neonatal face, pose estimation (-45°, 45°), and facial features (nose, mouth, and eyes) recognition. The NFAR approach exhibits a sensitivity, accuracy, and specificity of 98.7%, 98.5%, and, 95.7% respectively, for the newborn babies at the frontal (0°) facial region. The neonatal face and its attributes recognition can be expected to detect neonate's medical abnormalities unobtrusively by examining the variation in newborn facial texture pattern.

In dynamic and complicated industrial environments, feature-based SLAM based on laser scanner is a popular choice to achieve localization of autonomous guided vehicles. However, there are many clutters and dynamic objects degrading SLAM performance. This paper proposes a clutter-resistant SLAM solution where both point features generated from reflectors and line features are employed to improve SLAM robustness. First, a point feature recognition method based on geometrical characteristics of reflectors is developed to filter out clutters and identify true reflector landmarks; Then a dual-map based map management scheme is proposed for EKF-SLAM to further eliminate both types of fallacious landmarks and enhance its clutter resistance capability. The proposed methods eliminate adverse impact of clutters and thus improve SLAM performance in terms of accuracy, consistency and efficiency. The effectiveness of the proposed clutter-resistant SLAM solution is validated through real-time experiments. The absolute localization error is controlled within 19 mm and 31mm in X-axis and Y-axis respectively. The improved SLAM algorithm is proved to be accurate and efficient enough for practical application in dynamic and complicated industrial environments.

The ability to extract image features largely determines the accuracy of image classification. However, external interferences in images such as translation, rotation, scaling, occlusion, light, and non-linear deformation, result in greater intra-class differences and inter-class similarity, which substantially increases the difficulty of image classification. Benefitting from the excellent ability of feature learning and extraction, Convolutional Neural Networks (CNNs) have achieved good results in the field of image classification. However, they are also characterized by a complex training process and high-demand parameter adjustment. This paper propose a convolutional neural network optimization method to optimize the model parameters with hierarchical attribute constraints and achieve intelligent recognition of specific image features. Based on the optimized model, we further construct weak supervision and label correlation optimization models and provide a visualization solution for feature recognition results. Experimental results demonstrated that the proposed algorithm can efficiently realize intelligent image feature recognition with high accuracy.

This paper presents an effective multi-sensor approach for improving the accuracy of laser scanner by using a tactile probe to perform rapid and accurate reverse engineering of complex objects. The proposed system is unique in that it includes not only the physical integration of the two digitizers but also their combination at the measurement information level. With the coordinate data acquired using the optical scanner, intelligent data segmentation and feature recognition algorithm are proposed to divide the original point set into geometric elements and free-form surfaces. The tactile probe is guided to re-measure each feature with a small number of sampling points. Then different data fusion algorithms are proposed to compensate the scanned data patches of geometric features and free-form features with the accurate tactile probing data. Finally, the compensated point data can be exploited for accurate reverse engineering of a CAD model. Through the fusion of different sensor's information, both sensors complement each other with their advantages. Experimental results show that for the measurement of geometric surface, the proposed method can improve the accuracy of optical measurement to the accuracy of contact measurement. For free-form surface measurement, the proposed method reduces the optical measurement error from +0.032/-0.019 mm to ±0.016 mm, and the standard deviation from 0.012 mm to 0.004 mm. The measurement efficiency of proposed method is 4.5 times higher than that of contact measurement alone.

The identification of the characteristics of urban road traffic accidents is of great significance for reducing traffic accidents and the corresponding losses. In the context of big data, to accurately understand the characteristics of traffic accidents, the feature set of urban road traffic accidents is proposed, the XGBoost model is used to classify traffic accidents into minor accidents, general accidents, major accidents and serious accidents, and a GA-XGBoost feature recognition model is built. The GA-XGBoost feature recognition model is based on the genetic algorithm (GA) as a factor search algorithm and is verified by applying the big data of traffic accidents in a Chinese city from 2006 to 2016; in addition, the model is compared with the GA-RF, GA-GBDT and GA-LightGBM models. The results show that the GA-XGBoost model can accurately identify the features of the traffic accidents in 7 cities, including driving experience, illegal driving behavior, vehicle age, road intersection type, weather conditions, traffic flow and time interval. Compared with the GA-RF, GA-GBDT and GA-LightGBM models, the recognition features are more accurate, and the performance is better.

With the development of society, intelligent urban construction is drawing attention globally. It embed sensors and equipment into various environmental monitoring target objects to achieve environmental management and decision-making wisdom in a more granular and dynamic manner. However, how to achieve target objects' recognition in the dynamic environment is of essential importance for intelligent urban construction. Due to the shape, color and other characteristics for target objects are more similar, which make it difficult to identify the target types based on the low-level features such as color, shape, etc. In this paper, we attempt to apply the deep neural network composed of sparse autoencoders based unsupervised feature learning to identify the various types of target objects. On the other hand, due to the fact that the quantities of target objects which are obtained by environmental monitor may be not sufficient, cannot get more high-level visual feature through feature training, which affects the accuracy of the subsequent target recognition. A cross-domain feature learning scheme for target objects recognition using convolutional sparse auto-encoder has been presented. In order to improve the recognition speed, feature weights selection method based on a correlation analysis is further proposed for the purpose of reducing the amount of global features which are taken from target-domain images. Experimental results show that compared with non-transfer feature learning algorithm and underlying visual feature recognition algorithm, the new algorithm proposed in this paper has higher accuracy and robustness. Feature selection can reduce the computational time of global feature extraction and recognition by about 30% while improving recognition performance.

Text features in topographic maps are important for helping users to locate the area that a map covers and to understand the map's content. Previous works on the optical detection of map text from topographic maps have used geometric features, the Hough transform, and segmentation. However, these approaches still face challenges when detecting map text in complicated contexts, especially when the map text is touching other map features, such as contours or geographical features. Thus, state-of-the-art techniques for map text and feature recognition and manual interpretation and correction are always required to produce accurate results when optically converting topographic maps into a readable format. This paper proposes a methodological framework called the intelligent map reader that enables the automatic and accurate optical understanding of the content of a topographic map using deep learning techniques in combination with a gazetteer. The intelligent map reader framework includes the detection of map text via deep learning, the separation of text units via graph-based segmentation and clustering, optical character recognition (OCR) via an OCR engine, and digital-gazetteer-based map content understanding. Experimental results validate the efficiency and robustness of our proposed methodology for map text recognition and map content understanding. We expect the proposed intelligent map reader to contribute to various applications in the GeoAI field.

Subcutaneous veins should be accurately segmented to satisfy the emerging demands in biometrics-based identity recognition, automated intravenous injection, and vascular surgical navigation. This paper proposes a novel segmentation method for extracting vascular vein structures from near-infrared images based on multiple feature classification. First, the isotropic undecimated wavelet transform in combination with the Hessian function is used to enhance the visibility of vascular structures. An approximated vascular structure is extracted by multiple-scale features based on the enhanced image. The establishment of the mapping relationship between the sketch image and source image spaces is assisted by mending and fitting procedures. Vascular features are also extracted from the enhanced contour image by a matched filter. Furthermore, the image can be classified into vascular, skin, and blurred regions by a K-means algorithm. The final vascular segmentation outcome is a refined structure. Segmentation results demonstrated that the proposed method is robust for subcutaneous vein images with different blur scales, where the completeness and smoothness of the extracted vein structure are retained. The proposed method is accurate and completely automatic, with potential applications in biometric feature recognition and intravenous injection.

The challenges of selecting appropriate image features, optimizing complex nonlinear computations, and minimizing the approximation errors always exist in visual servoing. A fuzzy neural network controller is developed for a six-degrees-of-freedom robot manipulator to perform visual servoing is proposed to tackle these problems. To increase the accuracy of the image preprocesses, a synthetic image process performs feature extraction for the controller. The method combines a support vector machine contour recognition algorithm and a color-based feature recognition algorithm. For visual servoing, a control method based on the fuzzy cerebellar model articulation controller with the Takagi-Sugeno framework is proposed to directly map an image feature error vector to a desired robot end-effector velocity. This approach achieves visual servoing control without the need of computing the inverse interaction matrix. The control variables are learned and updated by the T-S fuzzy inference. This simplifies the implementation of visual servoing in real-time applications. The proposed control method is used to control an articulated manipulator with an eye-in-hand configuration. The results of simulations and experiments demonstrate that the proposed visual servoing controller has good performance, in terms of stability and convergence.

The purpose of this paper is to survey and assess the state-of-the-art in automatic target recognition for synthetic aperture radar imagery (SAR-ATR). The aim is not to develop an exhaustive survey of the voluminous literature, but rather to capture in one place the various approaches for implementing the SAR-ATR system. This paper is meant to be as self-contained as possible, and it approaches the SAR-ATR problem from a holistic end-to-end perspective. A brief overview for the breadth of the SAR-ATR challenges is conducted. This is couched in terms of a single-channel SAR, and it is extendable to multi-channel SAR systems. Stages pertinent to the basic SAR-ATR system structure are defined, and the motivations of the requirements and constraints on the system constituents are addressed. For each stage in the SAR-ATR processing chain, a taxonomization methodology for surveying the numerous methods published in the open literature is proposed. Carefully selected works from the literature are presented under the taxa proposed. Novel comparisons, discussions, and comments are pinpointed throughout this paper. A two-fold benchmarking scheme for evaluating existing SAR-ATR systems and motivating new system designs is proposed. The scheme is applied to the works surveyed in this paper. Finally, a discussion is presented in which various interrelated issues, such as standard operating conditions, extended operating conditions, and target-model design, are addressed. This paper is a contribution toward fulfilling an objective of end-to-end SAR-ATR system design.

Under a IEEE Open Access Publishing Agreement., Due to the technological evolution and the increasing popularity of smartphones, people can access an application using authentication based on biometric approaches from many different devices. Device interoperability is a very challenging problem for biometrics, which needs to be further studied. In this paper, we focus on interoperability device compensation for online signature verification since this biometric trait is gaining a significant interest in banking and commercial sector in the last years. The proposed approach is based on two main stages. The first one is a preprocessing stage where data acquired from different devices are processed in order to normalize the signals in similar ranges. The second one is based on feature selection taking into account the device interoperability case, in order to select to select features which are robust in these conditions. This proposed approach has been successfully applied in a similar way to two common system approaches in online signature verification, i.e., a global features-based system and a time functions-based system. Experiments are carried out using Biosecure DS2 (Wacom device) and DS3 (Personal Digital Assistant mobile device) dynamic signature data sets which take into account multisession and two different scenarios emulating real operation conditions. The performance of the proposed global features-based and time functions-based systems applying the two main stages considered in this paper have provided an average relative improvement of performance of 60.3% and 26.5% Equal Error Rate (EER), respectively, for random forgeries cases, compared with baseline systems. Finally, a fusion of the proposed systems has achieved a further significant improvement for the device interoperability problem, especially for skilled forgeries. In this case, the proposed fusion system has achieved an average relative improvement of 27.7% EER compared with the best performance of time functions-based system. These results prove the robustness of the proposed approach and open the door for future works using devices as smartphones or tablets, commonly used nowadays., This work was supported in part by the Project Bio-Shield under Grant TEC2012-34881, in part by Cecabank e-BioFirma Contract, and in part by Catedra UAM-Telefonica