Hongjun Li

• PHD
• PhD Student at Concordia University

Multi-object tracking is a crucial task in computer vision, but it faces significant challenges due to complex scenes and interactions between tracked objects. To address these challenges, we propose a complex movement-based spatial attribute tracking model (SACM). First, to enhance target detection, the model extracts both global and local feature...

Like deep neural network (DNN)-based classifiers, DNN-based trackers are also vulnerable to adversarial attacks that degrade the tracking performance by adding adversarial perturbations to the input videos. This paper proposes a detection method for the first time to assist the tracker in detecting adversarial attacks. The adversarial perturbations...

Future frame prediction for anomaly detection methods based on memory networks have been extensively explored in the academic domain. Nevertheless, traditional memory-guided network techniques, which store dispersed spatial low-dimensional features, often fall short in delivering satisfactory results when applied to datasets characterized by variab...

Due to the fuzziness of anomaly definition and the complexity of scenes in real video data, video anomaly detection is still a challenging task. In this work, we explored a novel lightweight dual branch convolution neural architecture that can separate appearance-motion representations to capture spatial and temporal information, respectively, sinc...

Video anomaly detection is a very challenging task in the field of computer vision. In this paper, we propose a multidimensional two-stage memory-guided network based on future frame prediction. Firstly, based on the understanding of the temporal correlation information between continuous video frames, we use an implicit computational approach to e...

The task of anomaly detection in surveillance videos is challenging due to the sparsity and diversity. In order to perceive more discriminative features and further improve performance, a grey-adversary perceptual network is proposed for anomaly detection. Our method is designed as a combination of frame prediction stage and frame optimization stag...

Combining Unmanned aerial vehicle (UAV) with artificial intelligence can effectively extract information as UAV fly flexibly and have a wide view, but the current processing efficiency of the video information acquired by UAV is low, resulting in insufficient use of resources. In order to enable the tracking algorithm to adapt to wide-angle target...

As a significant research hotspot in the field of computer vision, video anomaly detection plays an essential role in ensuring public safety. Anomaly detection remains a challenging task given the complex situation in public areas and the large random distribution of crowds. The density of people in the same scene varies greatly due to the instabil...

In this paper, we present a Heterogeneous Network with Multi-United-Memory (HN-MUM) module, which integrates motion and appearance to solve the Video Anomaly Detection (VAD) problem. First, we present a heterogeneous dual-flow network to process the motion and appearance information independently based on the notion of “specific analysis of particu...

With the popularity of surveillance equipment and the rise of intelligent surveillance, video anomaly detection has gradually become a research hotspot. Among them, for video processing, the three-channel video frame data can be directly used as the input of model, or some motion information can be extracted from the video frame, such as calculatin...

With the popularity of smart surveillance devices and the increase of people’s security awareness, video anomaly detection has become an important task. However, learning rich multi-scale spatio-temporal information from high-dimensional videos to predict anomalous behaviors is a challenging task due to the large local redundancy and complex global...

Video anomaly detection aims to identify the anomalous that do not conform to normal behavior. The abnormal events tend to relate to appearance and motion, in which there are considerable difference in each other. From the perspective of philosophy, “act according to circumstances”, we propose a dual-flow network to dissociate appearance informatio...

In order to integrate the ability of feature extraction of deep structure and short training time of broad structure, we propose a novel Vertical-Cross-Horizontal Network (VCHN) for data recognition, which mainly contains vertical operation, horizontal operation, nonlinear mapping and recognition decision. For vertical operation, we design a hierar...

Anomaly detection plays an important role in intelligent surveillance and has attracted increasing attention from researchers in recent years. It is generally regarded as discrimination that cannot be properly represented in most approaches. Despite its importance in probing the scarcity and indefinability of abnormal data during training, designin...

As a challenging visual task, visual object tracking has recently been composed of the classification and regression subtasks. The anchor-free regression network gets rid of the dependence on the anchors, but the redundant range makes it usually regress some samples involving non-target information. Evenly dividing a target by the regular receptive...

Neuro-fuzzy models have been applied in various domains, in which the issue of long time-consumption for optimizing parameters and less innovation in fuzzy method for feature extraction remains to be solved. Here, we present a novel cycle reinforce hierarchical model (CRHM) for effective and efficient recognition. The innovative strategies of CRHM...

Fall detection is drawing more attention from both academia and industry. The human body occupies smaller space relative to the background in images, so the complex background affects the extraction of human fall or non-fall features. In order to reduce the interference of the complex background, a fall detection method based on fused saliency maps...

Anomaly detection is an important and demanding problem in social harmony. However, due to the uncertainty, irregularity, diversity and scarcity of abnormal samples, the performance is often poor. This paper presents Generative Adversarial-Synergetic Networks (GA-SN) to improve the discriminative ability. It is built on the Adversarial Discriminant...

Speech not only conveys the content information but also reveals the emotions of speakers. In order to achieve effective speech emotion recognition, a novel multi-features integration algorithm has been proposed. The statistical Mel frequency cepstrum coefficient (MFCC) features are directly evolved from the original speech. To further mine more us...

In view of the fact that the current networks mostly improve the network performance and robustness by proposing multiple strategies to update the network parameters, this paper puts forward a novel task-driven transformed network idea. With optimizing network parameters, the network structure can be optimized by vertical update, horizontal update...

With the development of society and the progress of technology, more and more ocean activities are carried out. It results in booming of deep-sea diving. The use of helium-oxygen mixture (as a kind of breathing gas) solves the physiological problems of divers in saturated diving, but it brings about the Heliumspeech voice communication problem, the...

Due to the disadvantages that complex network structure, time-consuming training process and the insufficient feature extraction ability existing in deep structures and broad structures, Reinforce Cycle Cascade Model (RCCM) is proposed to offer a novel algorithm for image recognition. In RCCM, the cycle cascade model is used to extract the discrimi...

Human action recognition is a hot topic and it has been applied to various fields. Deep learning is one of the techniques in human action recognition which has achieved good results. However, the task is still challenging due to the less collected samples. In order to address this challenge and improve the recognition accuracy, the stepwise generat...

The most typical methods of human action recognition in videos rely on features extracted by deep neural network. Inspired by the temporal segment network, the sparse-temporal segment network to recognize human actions is proposed. Considering the sparse features contains the information of moving objects in videos, for example marginal information...

Data mining has actively contributed to solving many real-world problems with a variety of techniques. Traditional approaches in this field are classification, clustering and regression. During the last few years a number of chal-lenges have emerged, such as imbalanced data, multi-label and multi-instance problems, low quality and/or noisy data or...

The development of modern science and technology has seriously changed people's living and working habits. Bad sitting habit undoubtedly has an important impact on human health. Therefore, this paper proposes a real-time sitting posture recognition algorithm based on index map and BLS model. Firstly, use Kinect to collect body index maps and build...

Vehicle type recognition has become critical in intelligent transportation systems. The existing technology about vehicle type recognition is difficult to balance the recognition accuracy and recognition speed. Aiming at the problem of vehicle type recognition in the highway environment, a layered broad model combining the shallow feature layer wit...

Compressive tracking algorithm is a simple and efficient tracking algorithm, which has good accuracy and robustness performance. Despite much success has been demonstrated, numerous issues remain to be addressed. First, the samples selected by the algorithm do not have discriminative representation and the number of samples is relatively huge, affe...

Fall detection is a hot topic in the context of aging society, which can reduce the fall-related injuries effectively. In this paper, a fall detection algorithm based on visual saliency via deep feature learning is proposed. Firstly, we use the Kinect to collect and label RGB images and depth images of fall events. Secondly, the 2-stream convolutio...

Sparse representation is a hot issue in image processing domain while the dictionary learning is the key part of sparse representation. In order to improve the performance of sparse representation and solve the dictionary atoms selecting problem in the dictionary learning process for further image denoising, in this paper, a kind of robust and adap...

Fall recognition plays an important role in the life of the elderly. In this paper, a method of jointing object region location and deep feature learning for fall recognition is proposed. Firstly, we collect and analyze the color images and the depth images. Secondly, automatic location for object region on collected images is performed. Thirdly, w...

To improve the ability of image sparse representation of wavelet transform, an image compression algorithm based on the grey relation in the wavelet domain is proposed. First, the test image is decomposed by wavelet transform, and the wavelet coefficient in each scale is obtained. Then, based on the characteristics of the wavelet coefficients, the...

As the compressive tracking algorithm is easily failed to track target due to the classifier learning rate is fixed when appearances and lightings of target get seriously changed and completely loses the target after heavy occlusion, we propose adaptive compressive tracking algorithm based on perceptual hash algorithm. It makes the compressive trac...

No-Reference image quality assessment aims to predict the visual quality of distorted images without examining the original image as a reference. Most No-Reference image quality metrics which have been already proposed are designed for one or a set of predefined specific distortion types and are unlikely to generalize for evaluating images degraded...

Recently, robust sparse coding achieves high recognition rates on face recognition, even when dealing with occluded images. However, robust sparse coding is known that only guarantee the coefficient is global sparse when solving the sparse coefficient. In this paper, we divided the coefficient vector into multiple regions. Then, we enabled the elem...

Face database! (incomplete) Our database contains more than 1000 people, each people have 5-10 images. (differnt emotion,pose,etc.) Public release v1.0(10 December 2015) only contain 152 people! ------------------------------------------------------------------- Copyright (c) Nantong University. All rights reserved. This face database be used f...

Face recognition is an active topic in recognition systems, while the face occlusion is one of the most challenging problems for recognition. Recently, robust sparse coding achieves the state-of-the-art performance, especially when dealing with occluded images. However, robust sparse coding is known that only guarantee the coefficient is global spa...

In order to alleviate the influence of illumination, pose, expression and occlusion variations in face recognition, in this paper, an effective face recognition method based on discriminative sparse representation is proposed. To solve the problem of these variations, we extract discriminative features which represent for each of the training image...

We proposed a simple and efficient iteratively reweighted algorithm to improve the recovery performance for image reconstruction from compressive sensing (CS). The numerical experiential results demonstrate that the new proposed method outperforms in image quality and computation complexity, compared with standard l1-minimization and other iterativ...

We proposed a simple and efficient iteratively reweighted algorithm to improve the recovery performance for image reconstruction from compressive sensing(CS). The numerical experiential results demonstrate that the new proposed method outperforms in image quality and computation complexity, compared with standard1 l -minimization and other iterativ...

Traditional wavelet transform needs to be improved and perfected in sparse representation. In this paper, we proposed an image compression algorithm based on grey relational theory in wavelet domain. We use the character of wavelet coefficients, and apply the grey relational theory in coefficients relational description, and then propose an image c...

Traditional redundant dictionary can not represent the characters of image feature self-adaptively. In order to overcome theses drawbacks, this paper presents an optimal image dictionary construction algorithm. The algorithm establishes a novel optimal principle, using grey relational theory to analyze the value in a redundant dictionary. First we...

In this paper, improvements to the Non Local Means (NL-Means) algorithm introduced by Buades et al. are presented. The filtering parameter is unclearly defined in the original NL-Means algorithm. In order to solve this problem, we calculated filtering parameter by the relation of noise variance, and then proposed a noise variance estimate method...

Purpose The paper aims to do some further research on grey relational analysis applied in wavelet transform, and proposed a grey relational threshold algorithm for image denoising. This study tries to suppress the noise while retaining the edges and important structures as much as possible. Design/methodology/approach The paper analyzed the charac...

Purpose – The paper aims to do some further research on grey relational analysis applied in wavelet transform, and proposed a grey relational threshold algorithm for image denoising. This study tries to suppress the noise while retaining the edges and important structures as much as possible. Design/methodology/approach – The paper analyzed the ch...

Reliable video delivery has become one of the crucial issues in wireless video communications. Based on different error resilient property of forward error correction (FEC and Wyner-Ziv (WZ), we propose a scheme which combines these two technologies. Experimental results demonstrate the scheme can possess their both advantages and have an improveme...

We propose a no-reference method for image quality assessment to be used to various types of image distortions. We assess the image using statistic information of natural scenes, and use the generalized Gaussian density model to fit the marginal distribution of wavelet coefficients. Degree of image distortion is measured with the parameter values i...

Reliable video delivery has become one of the crucial issues in wireless video communications. Based on different error resilient property of forward error correction (FEC）and Wyner-Ziv (WZ), we propose a scheme which combines these two technologies. Experimental results demonstrate the scheme can possess their both advantages and have an improveme...

This study mainly discussed the application of grey theory in the non-subsampled Contourlet domain. The new algorithm combined the excellent characteristics of the non-subsampled Contourlet transform and grey theory in image denoising. The total variation model is used first to modify the noised image in order to reduce the pseudo-Gibbs artifacts....

To estimate noise variance of Gaussian white noise, a new method by the character of wavelet coefficients distribution is proposed. The wavelet coefficients in each sub-band can be well modelized by a Generalized Gaussian Distribution (GGD) whose parameters (scale and shape parameters) can be used to estimate noise variance. The simulation results...

Automatic blood cell tracking and velocity measurement in microvessels is a crucial task in biomedical and physiological research. For the analysis of the motion of blood cells in microvessels, a commonly used method for blood cell tracking and velocity estimation is spatiotemporal image-based analysis. However, in the process of the spatiotemporal...

Purpose – The traditional total variation (TV) models in wavelet domain use thresholding directly in coefficients selection and show that Gibbs' phenomenon exists. However, the nonzero coefficient index set selected by hard thresholding techniques may not be the best choice to obtain the least oscillatory reconstructions near edges. This paper aims...

The total variation (TV) image denoising model has been investigated in this study. The traditional TV model was reviewed and a new algorithm of the total variation model was proposed. The balance point of the image feature and image noise was considered to set the first stop criterion in the imaging processing iteration step. On the basis of the f...

The total variation (TV) image denoising model has been investigated in this study. The traditional TV model was reviewed and a new algorithm of the total variation model was proposed. The balance point of the image feature and image noise was considered to set the first stop criterion in the imaging processing iteration step. On the basis of the f...

In this paper we analyzed directional filter and pyramid filter in Contourlet transform, confirmed the different choices of filters directly affect the image denoising result, so we constructed a kind of compactly supported biorthogonal filter based on human visual Characteristics and applied it to image denoising. The new Contourlet transform algo...

A method for infrared image denoising based on the adaptive threshold Nonsubsampled Contourlet Transform (AT-NSCT) was proposed in this paper. It analyzed the construction method on NSCT and proved that the method can be applied to infrared image denoising. The AT-NSCT algorithm was applied to some kinds of infrared images and compared by both in i...