International Journal of Pattern Recognition and Artificial Intelligence

Due to rapid advances in the development of surveillance cameras with high sampling rates, low cost, small size and high resolution, video-based action recognition systems have become more commonly used in various computer vision applications. Human operators can be supported with the aid of such systems to detect events of interest in video sequences, improving recognition results and reducing failure cases. In this work, we propose and evaluate a method to learn two-dimensional (2D) representations from video sequences based on an autoencoder framework. Spatial and temporal information is explored through a multi-stream convolutional neural network in the context of human action recognition. Experimental results on the challenging UCF101 and HMDB51 datasets demonstrate that our representation is capable of achieving competitive accuracy rates when compared to other approaches available in the literature.

With the rapid growth of image collections, image classification and annotation has been active areas of research with notable recent progress. Bag-of-Visual-Words (BoVW) model, which relies on building visual vocabulary, has been widely used in this area. Recently, attention has been shifted to the use of advanced architectures which are characterized by multi-level processing. Hierarchical Max-Pooling (HMAX) model has attracted a great deal of attention in image classification. To improve image classification and annotation, several approaches based on ontologies have been proposed. However, image classification and annotation remain a challenging problem due to many related issues like the problem of ambiguity between classes. This problem can affect the quality of both classification and annotation results. In this paper, we propose an ontology-based image classification and annotation approach. Our contributions consist of the following: (1) exploiting ontological relationships between classes during both image classification and annotation processes; (2) combining the outputs of hypernym–hyponym classifiers to lead to a better discrimination between classes; and (3) annotating images by combining hypernym and hyponym classification results in order to improve image annotation and to reduce the ambiguous and inconsistent annotations. The aim is to improve image classification and annotation by using ontologies. Several strategies have been experimented, and the obtained results have shown that our proposal improves image classification and annotation.

Human emotion plays a major role in expressing their feelings through speech. Emotional speech recognition is an important research field in the human–computer interaction. Ultimately, the endowing machines that perceive the users’ emotions will enable a more intuitive and reliable interaction.The researchers presented many models to recognize the human emotion from the speech. One of the famous models is the Gaussian mixture model (GMM). Nevertheless, GMM may sometimes have one or more of its components as ill-conditioned or singular covariance matrices when the number of features is high and some features are correlated. In this research, a new system based on a weighted distance optimization (WDO) has been developed for recognizing the emotional speech. The main purpose of the WDO system (WDOS) is to address the GMM shortcomings and increase the recognition accuracy. We found that WDOS has achieved considerable success through a comparative study of all emotional states and the individual emotional state characteristics. WDOS has a superior performance accuracy of 86.03% for the Japanese language. It improves the Japanese emotion recognition accuracy by 18.43% compared with GMM and $k$-mean.

Automated recognition and classification of fishes are useful for studies dealing with counting of fishes for population assessments, discovering association between fishes and ecosystem, and monitoring of the ecosystem. This paper proposes a model which classifies the fishes belonging to the family Labridae in the genus and the species level. Features computed in the spatial and frequency domains are used in this work. All the images are preprocessed before feature extraction. Preprocessing step involves image segmentation for background elimination, de-noising and image enhancement. A combination of color, local binary pattern (LBP), histogram of oriented gradients (HOG), and wavelet features forms the feature vector. An ensemble feature reduction technique is used to reduce the attribute size. Performances of the system using combined as well as reduced feature sets are evaluated using seven popular classifiers. Among the classifiers, wavelet kernel extreme learning machine (ELM) showed higher classification accuracy of 96.65% in genus level and polynomial kernel ELM showed an accuracy of 92.42% in species level with the reduced feature set.

In this paper, we propose a robust variational segmentation model capable of overcoming the problem of the negative effects of outliers. The proposed method is based on the combination of the characteristics of the generalized mean with the concept of the active contours. The optimization problem raising in this combination employs a power method technique. We demonstrate the performance of the proposed model on a series of sample images from diverse modalities and show an outperforming proposed model in comparison with the state-of-the-art methods. The proposed method shows better or equivalent performance in terms of accuracy and robustness than the conventional state-of-the-art models. The validation of the efficiency of the proposed two-phase algorithm is further extended to vector-valued images and multi-phase formulation.

Virtual simulation and 3D interaction have shown great potentials in a variety of domains for our future life. For a virtual fish swarm simulation system, the simulation of cohesion behaviors of fish swarm and the interaction between human and fish swarm are two key components to create immersive interactive experiences. However, it is a huge challenge to create a realistic fish swarm simulation system while providing a natural and comfortable interaction. In this paper, we propose a method for immersive virtual fish swarm simulation based on infrared sensors. Based on dynamic weight constraints, we propose a particle swarm optimization method for fish swarm cohesion simulation, which separates a particle swarm by the state of each particle and dynamically controls the particle swarm, making the movement behavior of virtual fish more realistic. In addition, an interactive fast skinning method is proposed for cartoon fishes, which leverages image segmentation, Optical Character Recognition (OCR) and bone skinning are used to generate cartoon fishes based on user-created colors. With infrared sensors, we propose a method for virtual fish swarm interaction, where the positions of human skeleton are processed by an action analyzer, achieving real-time user interactions with fish swarms. With all the proposed techniques integrated in a system, the experimental results show that our method is feasible and effective.

Due to the limitation of industrial and civil fields of the polarization camera, we have developed a scheme for synthesizing polarization image based on intensity image and simulation technology. A whole process polarization or intensity transfer model from the incident solar light to the point at which the camera enters the pupil is set up based on 6SV. The indoor simulation test system of the target surface reflectivity is formed. The simulated polarization image is compared with the image directly obtained by the polarization camera under the same condition. Compared with the image from the polarization camera, we get the same trend of contrast information, while the target characteristic information is obviously improved for the intensity image and the error rate of the improved contrast does not exceed 12%.

Aimed at scene matching problem for taking infrared image as the actual data and the visible image as the referenced data, a multi-resolution matching algorithm which fuses compressive sensing Scale Invariant Feature Transform (SIFT) feature is presented based on Bandelet transform. Two kinds of images are separately transformed into Bandelet domain to compress the feature search space of scene matching based on the best sparse representation of natural images by Bandelet transform. On the basis of adaptive Bayes threshold denoising for infrared image, the concept of sparse feature representation of compressive sensing theory is introduced into SIFT algorithm. For low-frequency image in Bandelet domain, high-dimensional SIFT key point feature description vector is projected on compressive sensing random measurement matrix to achieve dimensionality reduction. Then, the improved Genetic Algorithm (GA) to overcome premature phenomena is used as the search strategy, and the L₁ distance measure of SIFT feature vectors of compressive sensing for two kinds of images is applied to the search similarity criterion to match low-frequency image of high scale in Bandelet domain. The matching result is used as the guidance of the matching process for low-frequency image of low scale, and the matching result of full-resolution image is obtained iteratively. Experimental results show that the proposed method has not only high matching accuracy and fast matching speed, but also better robustness in comparison with some classic matching algorithms, which can resist the geometric distortion of rotation for actual image.

Threshold segmentation has been widely used in recent years due to its simplicity and efficiency. The method of segmenting images by the two-dimensional maximum entropy is a species of the useful technique of threshold segmentation. However, the efficiency and stability of this technique are still not ideal and the traditional search algorithm cannot meet the needs of engineering problems. To mitigate the above problem, swarm intelligent optimization algorithms have been employed in this field for searching the optimal threshold vector. An effective technique of lightning attachment procedure optimization (LAPO) algorithm based on a two-dimensional maximum entropy criterion is offered in this paper, and besides, a chaotic strategy is embedded into LAPO to develop a new algorithm named CLAPO. In order to confirm the benefits of the method proposed in this paper, the other seven kinds of competitive algorithms, such as Ant–lion Optimizer (ALO) and Grasshopper Optimization Algorithm (GOA), are compared. Experiments are conducted on four different kinds of images and the simulation results are presented in several indexes (such as computational time, maximum fitness, average fitness, variance of fitness and other indexes) at different threshold levels for each test image. By scrutinizing the results of the experiment, the superiority of the introduced method is demonstrated, which can meet the needs of image segmentation excellently.

Hospitals have accumulated a large amount of medical image data which need to be analyzed and integrated so as to be able to find the needed medical image in time, which is the basis of key technologies such as intelligent diagnosis of diseases. Meanwhile, through the analysis and integrated processing of medical images, the potential value of existing medical image data can be fully explored. In this paper, the key technologies in the intelligent image knowledge discovery system and the characteristics of medical image data are studied and improved.

In this paper, the characteristics of knowledge discovery and medical image data are comprehensively considered, and RDM texture features are selected as the feature representation of medical images. An improved RDM operator is proposed and proved by experimental results. Experimental results show that the improved RDM coding method can improve the stability of medical image data expression.

Task scheduling between edge devices and remote servers is a common application scenario in edge computing or cloud computing, also known as computational offloading. A reasonable scheduling strategy can effectively shorten task completion time, reduce energy consumption, and improve user experience. However, the traditional offline task scheduling algorithm is NP-hard, and the decision requires obtaining all the information of the task and the device (such as task computing amount, data amount, device computing resources, etc.), which is challenging to meet in practical applications. The semi-online algorithm describes the task scheduling method when the system cannot obtain all the information. In this paper, we propose an Efficient Semi-online algorithm for Multi-users task offloading (ESaM), which includes two specific implementations: ESaM-I as known server-side idle time, and ESaM-O for known task computing amount. Because ESaM-I has obtained server information, it is better than ESaM-O in performance for most of the scenarios. The experimental results show that ESaM-I and ESaM-O are superior to the well-known semi-online scheduling algorithm SPaC in task completion time. As the remote processor computing ability increases, the average makespan converges to 0.875, 0.742, 0.782 for SPaC-M, ESaM-O, and ESaM-I in the simulation.

Most nonstationary and time-varying mixed source separation algorithms are based on the model of instantaneous mixtures. However, the observation signal is a convolutional mixed source in reverberation environment, such as mobile voice received by indoor microphone arrays. In this paper, a time-varying convolution blind source separation (BSS) algorithm for nonstationary signals is proposed, which can separate both time-varying instantaneous mixtures and time-varying convolution mixtures. We employ the variational Bayesian (VB) inference method with Gaussian process (GP) prior for separating the nonstationary source frame by frame from the time-varying convolution signal, in which the prior information of the mixing matrix and the source signal are obtained by the Gaussian autoregressive method, and the posterior distributions of parameters (source signal and mixing matrix) are obtained by the VB learning. In the learning process, the learned parameters and hyperparameters are propagated to the next frame for VB inference as the prior which is combined with the likelihood function to get the posterior distribution. The experimental results show that the proposed algorithm is effective for separating time-varying mixed speech signals.

Online transaction log records the relevant information of the users, commodities and transactions, as well as changes over time, which can help analysts understand commodities’ sales. The existing visualization methods mainly analyze the purchase behavior from the perspective of users, while analyzing the sales trend of commodities can better help merchants to make business decisions. Based on the transaction log, this paper puts forward the visual analysis framework of commodity sales trend and the corresponding data processing algorithm. The concepts of volatility and dynamic performance of sales trend are proposed, through which the multi-dimensional sales data of time-oriented are displayed in two-dimensional space. The “Feature Ring” is designed to display the detailed sales information of the products. Based on the above methods, a visual analysis system is designed and implemented. The usability and validity of the visualization methods are verified by using JD online transaction data. The visualization methods enable manufacturers to formulate production plans and carry out product research and develop better.

Aiming at the detection and tracking of moving targets in industrial automation system, a dynamic target tracking algorithm based on HAAR and CAMSHIFT is proposed. A cascade HAAR classifier is designed and trained for tracking targets. CAMSHIFT algorithm is used to track and detect moving targets quickly. The system is tested on Raspberry Pi embedded platform. The results show that the algorithm can detect the target correctly and track the target effectively.

Natural disasters have had great impact on human beings. Emergency relief is becoming more important with the increase of the frequency and scale of humanitarian emergencies resulting from more natural disasters because of climate change. In this paper, an interesting hybrid knowledge representation method during the development of a KBS for design of emergency relief structures is presented. It encapsulates ill-structured, semi-structured and structured knowledge that is gathered from literature, human expert and even knowledge gleaned during the system development. All routine as well as cumbrous activities in the emergency relief cycle are covered. The system can provide the user with advice on preliminary plan evaluation, plan optimization, plan evaluation, plan summary and miscellaneous. It would be beneficial to the field of disaster emergency relief decision by focusing on the acquisition and organization of expert knowledge through the development of knowledge-based system.

It is known that some of the algorithms in optimization field have originated from inspiration from animal behaviors in nature. Natural phenomena such as searching behavior of ants for food in a collective way, movements of birds and fish groups as swarms provided the inspiration for solutions of optimization problems. Traveling Salesman Problem (TSP), a classical problem of combinatorial optimization, has implementations in planning, scheduling and various scientific and engineering fields. Ant colony optimization (ACO) and Particle swarm optimization (PSO) techniques have been commonly used for TSP solutions. The aim of this paper is to propose a new hierarchical ACO- and PSO-based method for TSP solutions. Enhancing neighboring operators were used to achieve better results by hierarchical method. The performance of the proposed system was tested in experiments for selected TSPLIB benchmarks. It was shown that usage of ACO and PSO methods in hierarchical structure with neighboring operators resulted in better results than standard algorithms of ACO and PSO and hierarchical methods in literature.