Occlusion Sequence Mining for Activity Discovery from Surveillance Videos

Complex multiobject interactions result in occlusion sequences, which are a visual signature for the event. In this work, multiobject interactions are tracked using a set of qualitative occlusion primitives derived on the basis of the persistence hypothesis: objects continue to exist even when hidden from view. Variable length temporal sequences of occlusion primitives are shown to be well correlated with many classes of semantically significant events. In surveillance applications, determining occlusion primitives is based on foreground blob tracking and requires no prior knowledge of the domain or camera calibration. New foreground blobs are identified as putative objects that may undergo occlusions, split into multiple objects, merge back again, and so forth. Significant activities are identified through temporal sequence mining; these bear high correlation with semantic categories (e.g., disembarking from a vehicle involves a series of splits). Thus, semantically significant event categories can be recognized without assuming camera calibration or any environment/object/action model priors.

An Automatic Off-Line Signature Verification and Forgery Detection System

This chapter presents an off-line signature verification and forgery detection system based on fuzzy modeling. The various handwritten signature characteristics and features are first studied and encapsulated to devise a robust verification system. The verification of genuine signatures and detection of forgeries is achieved via angle features extracted using a grid method. The derived features are fuzzified by an exponential membership function, which is modified to include two structural parameters. The structural parameters are devised to take account of possible variations due to handwriting styles and to reflect other factors affecting the scripting of a signature. The efficacy of the proposed system is tested on a large database of signatures comprising more than 1,200 signature images obtained from 40 volunteers.

State of the Art in Off-Line Signature Verification

Automatic signature verification is a biometric method that can be applied in all situations where handwritten signatures are used, such as cashing a check, signing a credit card, authenticating a document, and others. Over the last two decades, several innovative approaches for off-line signature verification have been introduced in literature. Therefore, this chapter presents a survey of the most important techniques used for feature extraction and verification in this field. The chapter also presents strategies used to face the problem of a limited amount of data, as well as important challenges and research directions.

Elastic Matching Techniques for Handwritten Character Recognition

This chapter reviews various elastic matching techniques for handwritten character recognition. Elastic matching is formulated as an optimization problem of planar matching, or pixel-to-pixel correspondence, between two character images under a certain matching model, such as affine and nonlinear. Use of elastic matching instead of rigid matching improves the robustness of recognition systems against geometric deformations in handwritten character images. In addition, the optimized matching represents the deformation of handwritten characters and thus is useful for statistical analysis of the deformation. This chapter argues the general property of elastic matching techniques and their classification by matching models and optimization strategies. It also argues various topics and future work related to elastic matching for emphasizing theoretical and practical importance of elastic matching.

A Recurrent Probabilistic Neural Network for EMG Pattern Recognition

In the field of pattern recognition, probabilistic neural networks (PNNs) have been proven as an important classifier. For pattern recognition of EMG signals, the characteristics usually used are: (1) amplitude, (2) frequency, and (3) space. However, significant temporal characteristic exists in the transient and non-stationary EMG signals, which cannot be considered by traditional PNNs. In this article, a recurrent PNN, called recurrent log-linearized Gaussian mixture network (R-LLGMN), is introduced for EMG pattern recognition, with the emphasis on utilizing temporal characteristics. The structure of R-LLGMN is based on the algorithm of a hidden Markov model (HMM), which is a routinely used technique for modeling stochastic time series. Since R-LLGMN inherits advantages from both HMM and neural computation, it is expected to have higher representation ability and show better performance when dealing with time series like EMG signals. Experimental results show that R-LLGMN can achieve high discriminant accuracy in EMG pattern recognition.

Implementing Negative Correlation Learning in Evolutionary Ensembles with Suitable Speciation Techniques

Negative correlation learning (NCL) is a technique that attempts to create an ensemble of neural networks whose outputs are accurate but negatively correlated. The motivation for such a technique can be found in the bias-variance-covariance decomposition of an ensemble of learner’s generalization error. NCL is also increasingly used in conjunction with an evolutionary process, which gives rise to the possibility of adapting the structures of the networks at the same time as learning the weights. This chapter examines the motivation and characteristics of the NCL algorithm. Some recent work relating to the implementation of NCL in a single objective evolutionary framework for classification tasks is presented, and we examine the impact of two speciation techniques: implicit fitness sharing and an island model population structure. The choice of such speciation techniques can have a detrimental effect on the ability of NCL to produce accurate and diverse ensembles and should therefore be chosen carefully. This chapter also provides an overview of other researchers’ work with NCL and gives some promising future research directions.

A Novel Support Vector Machine with Class-Dependent Features for Biomedical Data

This chapter introduces an approach to class-dependent feature selection and a novel support vector machine (SVM). The relative background and theory are presented for describing the proposed method, and real applications of the method on several biomedical datasets are demonstrated in the end. The authors hope this chapter can provide readers a different view of feature selection method and also the classifier so as to promote more promising methods and applications.

Human Detection in Static Images

Human detection is the first step for a number of applications such as smart video surveillance, driving assistance systems, and intelligent digital content management. It’s a challenging problem due to the variance of illumination, color, scale, pose, and so forth. This chapter reviews various aspects of human detection in static images and focuses on learning-based methods that build classifiers using training samples. There are usually three modules for these methods: feature extraction, classifier design, and merge of overlapping detections. The chapter reviews most existing methods for each module and analyzes their respective pros and cons. The contribution includes two aspects: first, the performance of existing feature sets on human detection are compared; second, a fast human detection system based on histogram of oriented gradients features and cascaded AdaBoost classifier is proposed. This chapter should be useful for both algorithm researchers and system designers in the computer vision and pattern recognition community.

Robust Face Recognition Technique for a Real-Time Embedded Face Recognition System

In this chapter, we propose a pose variability compensation technique, which synthesizes realistic frontal face images from nonfrontal views. It is based on modeling the face via active appearance models and estimating the pose through a correlation model. The proposed technique is coupled with adaptive principal component analysis (APCA), which was previously shown to perform well in the presence of both lighting and expression variations. The proposed recognition techniques, though advanced, are not computationally intensive. So they are quite well suited to the embedded system environment. Indeed, the authors have implemented an early prototype of a face recognition module on a mobile camera phone so the camera can be used to identify the person holding the phone.

Seeking Patterns in the Forensic Analysis of Handwriting and Speech

This chapter examines features of handwriting and speech and their effectiveness at determining whether the identity of a writer or speaker can be identified from his or her handwriting or speech. For handwriting, some of the subjective and qualitative features used by document examiners are investigated in a scientific and quantitative manner based on the analysis of three characters (“d,” “y,” and “f”) and the grapheme “th.” For speech, several frequently used features are compared for their strengths and weaknesses in distinguishing speakers. The results show that some features do have good discriminative power, while others are less effective. Acceptable performance can be obtained in many situations using these features. However, the effect of handwriting forgery/disguise or conscious speech imitation/alteration on these features is not investigated. New and more powerful features are needed in the future if high accuracy person identification can be achieved in the presence of disguise or forgery.