Ba-Vui Le

Using weighted dynamic range for histogram equalization to improve the image contrast

In this paper, an effective method, named the brightness preserving weighted dynamic range histogram equalization (BPWDRHE), is proposed for contrast enhancement. Although histogram equalization (HE) is a universal method, it is not suitable for consumer electronic products because this method cannot preserve the overall brightness. Therefore, the output images have an unnatural looking and more visual artifacts. An extension of the approach based on the brightness preserving bi-histogram equalization method, the BPWDRHE used the weighted within-class variance as the novel algorithm in separating an original histogram. Unlike others using the average or the median of gray levels, the proposed method determined gray-scale values as break points based on the within-class variance to minimize the total squared error of each sub-histogram corresponding to the brightness shift when equalizing them independently. As a result, the contrast of both overall image and local details was enhanced adequately. The experimental results are presented and compared to other brightness preserving methods.

Interactive activity recognition using pose-based spatio-temporal relation features and four-level Pachinko Allocation Model

Novel interactive activity recognition method using a topic modeling technique.Pose-based spatio-temporal relation features for intra- and inter-person.Two types of codeword corresponding to joint distance and angle.Four-level Pachinko Topic Model (PAM) for flexibly modeling interactions.Outperformance of recognition accuracy to state-of-the-art methods. In this paper, we go beyond the problem of recognizing video-based human interactive activities. We propose a novel approach that permits to deeply understand complex person-person activities based on the knowledge coming from human pose analysis. The joint coordinates of interactive objects are first located by an efficient human pose estimation algorithm. The relation features consisting of the intra and inter-person features of joint distance and angle, are suggested to use for describing the relationships between body components of the individual persons and the interacting two participants in the spatio-temporal dimension. These features are then provided to the codebook construction process, in which two types of codeword are generated corresponding to distance and angle features. In order to explain the relationships between poses, a flexible hierarchical topic model constructed by four layers is proposed using the Pachinko Allocation Model. The model is able to represent the full correlation between the relation features of body components as codewords, the interactive poselets as subtopics, and the interactive activities as super topics. Discrimination of complex activities presenting similar postures is further obtained by the proposed model. We validate our interaction recognition method on two practical data sets, the BIT-Interaction data set and the UT-Interaction data set. The experimental results demonstrate that the proposed approach outperforms recent interaction recognition approaches in terms of recognition accuracy.

PAM-based flexible generative topic model for 3D interactive activity recognition

Interactive activity recognition from the RGB videos still remains a challenge, therefore some existing approaches paid the attention to RGB-Depth video process to avoid problems relating to mutual occlusion and redundant human pose and to improve accuracy of skeleton extraction. From the single action to complex interaction activity, it is necessary an efficient model to describe the relationship of body components between multi-human objects. In this research, the authors proposed a hierarchical model based on the Pachinko Allocation Model for interaction recognition. Concretely, the joint features comprising joint distant and joint motion are calculated from the skeleton position and then support to topic modeling. The probabilistic models describing the flexible relationship between features - poselets - activities are generated by this model. Finally, the Binary Tree of Support Vector Machine is applied for classification. Compared with existing state-of-the-arts, the proposed method outperforms in overall classification accuracy (8-21% approximately) with the SBU Kinect Interaction Dataset.

Traffic Behavior Recognition Using the Pachinko Allocation Model.

CCTV-based behavior recognition systems have gained considerable attention in recent years in the transportation surveillance domain for identifying unusual patterns, such as traffic jams, accidents, dangerous driving and other abnormal behaviors. In this paper, a novel approach for traffic behavior modeling is presented for video-based road surveillance. The proposed system combines the pachinko allocation model (PAM) and support vector machine (SVM) for a hierarchical representation and identification of traffic behavior. A background subtraction technique using Gaussian mixture models (GMMs) and an object tracking mechanism based on Kalman filters are utilized to firstly construct the object trajectories. Then, the sparse features comprising the locations and directions of the moving objects are modeled by PAM into traffic topics, namely activities and behaviors. As a key innovation, PAM captures not only the correlation among the activities, but also among the behaviors based on the arbitrary directed acyclic graph (DAG). The SVM classifier is then utilized on top to train and recognize the traffic activity and behavior. The proposed model shows more flexibility and greater expressive power than the commonly-used latent Dirichlet allocation (LDA) approach, leading to a higher recognition accuracy in the behavior classification.

Hierarchical emotion classification using genetic algorithms

Emotion classification from speech signal is an interesting subject of machine learning applications that can provide the emotional or psychological states from speakers. This implicit information is helpful for machine to understand human behavior in more comprehensive way. Many feature extraction and classification methods have being proposed to find the most accurate and efficient method, but this is still an open question for researchers. In this paper, we propose a novel method to select features and classify emotions in hierarchical way using genetic algorithm and support vector machine classifiers in order to find the most accurate binary classification tree. We show the efficiency and robustness of our method by applying and analyzing on Berlin dataset of emotional speech and the experiment results show that our method achieves high accuracy and efficiency.

Clustering based multi-object positioning system

Acoustic source positioning plays an important role in military tracking unwelcome objects. A system for this application must be capable of dealing with the input recorded convolved mixture signals while minimizing the high communication and computation cost. This paper describes a distributed system for positioning multiple independent moving sources relying on acoustic signals. The sensors pre-process the sensed data to obtain the frequency features before compressing and sending it to the base. At the base, the source positioning are carried out via two clustering stages and an optimization method. Analysis and simulation results show that our system provides high accuracy and needs neither much communication nor complex computation in a distributed manner. It is robust even when there exists high noise with Rayleigh multi-path fading under Doppler effect and when the number of independent sources is greater than the microphone number.

Describing body-pose feature - poselet - activity relationship using Pachinko Allocation Model

Understanding video-based activities have remained the challenge regardless of efforts from the image processing and artificial intelligence community. However, the rapid developing of computer vision in 3D area has brought an opportunity for the human pose estimation and so far for the activity recognition. In this research, the authors suggest an impressive approach for understanding daily life activities in the indoor using the skeleton information collected from the Microsoft Kinect device. The approach comprises two significant components as the contribution: the pose-based feature extraction under the spatio-temporal relation and the topic model based learning. For extracting feature, the distance between two articulated points and the angle between horizontal axis and joint vector are measured and normalized on each detected body. A codebook is then constructed using the K-means algorithm to encode the merged set of distance and angle. For modeling activities from sparse features, a hierarchical model developed on the Pachinko Allocation Model is proposed to describe the flexible relationship between features - poselets - activities in the temporal dimension. Finally, the activities are classified by using three different state-of-the-art machine learning techniques: Support Vector Machine, K-Nearest Neighbor, and Random Forest. In the experiment, the proposed approach is benchmarked and compared with existing methods in the overall classification accuracy.

Background subtraction with neighbor-based intensity correction algorithm

An efficient foreground detection algorithm is presented in this work to be robust against consecutively illuminance changes and noise, and adaptive with dynamic speeds of motion in the background. The scene background is firstly modeled by a novel algorithm, namely Neighbor-based Intensity Correction, which identifies and modifies motion pixels extracted from the difference of the background and the current frame. Concretely the first frame is assumed as an initial background to be updated at each new coming frame based on the mechanism of the standard deviation value comparison. Two pixel windows used for standard deviation calculation are generated surrounding a corresponding motion pixel from the background and the current frame. The steadiness of the current background at the pixel-level is measured by a constantly updating factor to decide the usage of the algorithm or not. In the next stage, the foreground of the current frame are detected by the background subtraction scheme with an optimal Otsu threshold. This method is evaluated on various well-known datasets in the object detection and tracking area and then compared with recent approaches via some common quantitative measurements. From experimental results, the proposed method achieves the better results (approximately 5-20%) in term of the foreground detection accuracy.

PAM-Based Behavior Modelling

A novel approach for human behavior modelling is represented in this paper based on the Pachinko Allocation Model (PAM) algorithm for the video-based road surveillance. In particular, the authors focus on the behavior analysis and modelling for learning and training as the main distribution of this research. Sparse object features in sequence of frames are modelled into activities and behaviors with full topic correlations to avoid omissions of small activities.