In this paper, a novel graph-based transductive classification approach, called Linear Neighborhood Propagation, is proposed. The basic idea is to predict the label of a data point according to its neighbors in a linear way. This method can be cast into a second-order intrinsic Gaussian Markov random field framework. Its result corresponds to a solution to an approximate inhomogeneous biharmonic equation with Dirichlet boundary conditions. Different from existing approaches, our approach provides a novel graph structure construction method by introducing multiple-wise edges instead of pairwise edges, and presents an effective scheme to estimate the weights for such multiple-wise edges. To the best of our knowledge, these two contributions are novel for semi-supervised classification. The experimental results on image segmentation and transductive classification demonstrate the effectiveness and efficiency of the proposed approach.

This paper presents a new method to enforce inverse consistency in nonrigid image registration and matching. Conventional approaches assume diffeomorphic transformation, implicitly or explicitly. However, the inherent smoothness constraint discourages discontinuity consideration. We propose a post-processing algorithm that integrates the input forward and backward fields, which are output by existing registration/matching algorithms, to produce more robust results. Given such a pair of input fields, our algorithm alternately refines the fields by tensor belief propagation, and enforces inverse consistency in stochastic sense by generalized total least squares fitting. To show the efficacy of our stochastic inverse consistency approach, we first present results on very noisy fields. We then demonstrate improvement on existing stereo matching where occlusion is naturally handled by localizing violations of inverse consistency. Finally, we propose a novel application on image stitching, where stochastic inverse consistency is employed in structure deformation, in order to seamlessly align overlapping images with severe misalignment in structure and intensity.

In this paper, we address the general problem of learning from both labeled and unlabeled data. Based on the reasonable assumption that the label of each data can be linearly reconstructed from its neighbors’ labels, we develop a novel approach, called Linear Neighborhood Propagation (LNP), to learn the linear construction weights and predict the labels of the unlabeled data from the labeled data. The main procedure is composed of two steps: (1) it computes the local neighborhood geometry according to the image information by solving a constrained least squares problem, (2) the obtained geometry will be used to predict the labels of unlabeled data by solving a linearly constrained quadratic optimization problem. LNP has at least three characteristics. Firstly, it is much practical in that only one free parameter, the size of neighborhood, requires user-tuning. Secondly it can be easily extended to out-of-sample examples. Finally the labels achieved from LNP can be sufficiently smooth with respect to the intrinsic structure collectively revealed by the labeled and unlabeled points. Many promising experimental results, such as object recognition, data ranking and interactive image segmentation, demonstrate the effectiveness and efficiency of the proposed approach.

For high-dimensional data classification problems such as face recognition, one of the most efficient classifiers is the nearest neighbor (NN) classifier. What mostly affects the NN classification performance is the feature extracted by some methods. And the kernel method is one of the efficient methods for extracting features. However, the selection of kernel parameters is still difficult. In this paper, we propose a so-called data dependent kernel (DDK) which is defined by generalizing the Gaussian kernel. Also an efficient and practical method is presented to calculate the DDK parameters. Moreover, one DDK based on subspaces is given to improve the recognition performance. Experiments show that the proposed DDK can achieve promising classification performance in face recognition and SPECT heart diagnosis.

This paper investigates a new approach for the palmprint identification using real Gabor function (RGF) filtering. Inkless composite hand images have been used to automatically to extract the palmprints from peg-free imaging setup. These palmprints, after normalization, are subjected to selective feature sampling by a bank of RGF. Each of these filtered images has been used to extract significant features (PalmCode) from each of 6 concentric circular bands. Our preliminary experimental results using 400 low-resolution palmprint images achieve the recognition rate of 97.50% and also illustrate the shortcomings of results presented in earlier work. The results show the uniqueness of palmprint texture, even in the two hands of an individual and its possible use in biometrics based personal recognition.

Investigates two methods for the detection of defects on textured surfaces using neural networks and support vector machines. Every pixel from the inspection image is characterized by a feature vector, which serves as a local measure of homogeneity of texture. The feature vectors from the gray-level arrangement of neighboring pixels are transformed to eigenspace using Principal Component Analysis (PCA). The transformed features from a predetermined set of training images are used to train the classifier. The trained classifier is used to classes every pixel from inspection image into two-class, i.e. with- or without-defect. The experimental results on real fabric defects show that the proposed scheme can successfully segment the defects from the inspection images.

The basic idea of RISC (reduced intricacy sensing and control) robotics is an attempt to perform challenging industrial manufacturing tasks by using a combination of simple hardware and sophisticated algorithms. Many effective strategies and algorithms have been explored. However, the issue of optimal design of RISC sensor has not been solved. The main reason is the shortage of good models. In this paper, we propose a statistical model for one of the typical RISC sensors, i.e. the crossbeam sensor. Based on this statistical model we employ the multiple hypotheses-testing method as optimal design technique and present its applied strategies. It is believed that this work will lead to development of new RISC sensors because a new principle and a pertinent model are introduced into this area.

We present a generalized multiple hypotheses testing strategy for a fully distributed multisensor system. Each sensor makes its own decision based on a set of rules which is derived by minimizing a Bayes cost risk function. The decision rules for each sensor are tightly coupled with other sensors since the "interdependence" between sensors is defined by the costs associated with each decision. For a special set of cost values, our proposed multiple hypotheses testing strategy becomes the common Bayes approach. We can also quantify the performance of the system in terms of the probabilities of correct, incorrect and uncertain decisions.

In this paper, we present a new multiple hypothesis testing (MHT) method for single sensor to make a decision. Generalizing the work of Wang, Shen and Qian (1998) from binary hypothesis testing to multiple hypothesis testing, we partition the decision space for multiple hypotheses problems and offer more decision options to increase decision flexibility. Decision rules are derived. The classical Bayes method for multiple hypothesis problem can be viewed as one special case. Performance measure of the sensor can also be computed.

We present a multiple hypothesis testing method for multisensor systems. Generalizing previous work by Wang et al. (1996, 1998) from binary hypothesis testing to multiple hypotheses testing, we re-partition the decision space for multiple hypotheses problems and offer a third decision option to increase the decision flexibility and reflect reality. The set of optimal decision rules is derived for multisensor systems with a fusion center. System performance can also be quantified. A numerical example is given to illustrate the proposed method.

We present computer vision techniques for building dressed human models using images. We develop an algorithm for three-dimensional body reconstruction and texture mapping using contour, stereo, and texture information from several images and deformable superquadrics as the model parts. We demonstrate a novel vision technique for analysis of cloth draping behavior. This technique allows for estimation of cloth model parameters, such as bending properties, but can also be used to estimate the contact points between the body and clothing in the range data of dressed humans. Combined with our body reconstruction algorithm and additional constraints on the articulation model, the detection of the garment-body contact points allows construction of a dressed human model in which even the geometry that was covered by clothing in the available data is reasonably well estimated.

In this paper we present a method for 3-D reconstruction of human bodies with application in CAD systems for garment design. The reconstruction scheme uses image information from several arbitrary views and deformable superquadrics as the models of the body parts. Two visual cues are used: occluding contours and stereo (possibly aided by projected patterns). Our preliminary experiments show that the reconstruction is more complete than in purely stereo or structured light based methods and more precise than the reconstruction from occluding contours only. From the reconstructed human body, the body measurements can be taken automatically, and used in garment design. We give an example of draping of virtual garment over the photo-realistic 3D model of the imaged human. One can easily envision the use of the described algorithms in the development of custom-fit garment retail software over the Internet, which would include the possibility of trying the garment on in virtual reality.

In sensory data fusion and integration consideration, sensor independence is a common assumption. We demonstrate the impact of including dependent information in the sensory data combination process. The team consensus approach based on information entropy can improve the measurement accuracy remarkably. The major benefits of the approach are: (a) the simple linear combination of the weighted initial local estimates for each sensor; and (b) the low order bivariate likelihood functions which can be represented easily. A comparison of the team consensus approach with the Bayesian approach is presented.

Presents a hypothesis testing method called double bound testing which is used for statistical decision-making, specifically for binary decisions. A probability decision space is defined to increase the decision flexibility. Based on the decision reached by each sensor, a combination rule is also formulated to give the global decision for the multisensor system. The proposed method offers three options for decision snaking, rather than the classical binary options. An experiment on 2D object identification was performed to demonstrate the proposed strategy.

It is common to assume sensor independence in the sensory data fusion and integration. The authors previously (1997, 1998) illustrated that the team consensus approach based on information entropy can remarkably improve the measurement accuracy. The major benefits of the approach are (a) the simple linear combination of the weighted initial expected estimates for each sensor; and (b) the low order bivariate likelihood functions which can be represented easily. In this paper, we demonstrate specifically both the positive and negative impacts of including dependent information in sensory data combination process; and show how the measurable consensus uncertainty level can be derived. A comparison of the team consensus approach with the Bayesian approach is presented.

In this paper, a decentralized approach based on the team consensus approach and Markovian model is proposed to integrate multisensory data. A team of sensors can estimate the local and global uncertainties utilizing self-entropy and conditional-entropy measures of the sensors. Consensus can be reached based on the initial expected values and "uncertainty" weights assigned by the sensors. The proposed approach is compared with the Bayesian approach via experiments on two independent sensors. Results showed that consensus reached are comparable. However, there are factors that indicated the decentralized approach requires less communication and computational effort to reach consensus among sensors.

The authors investigate the issue of information interdependence in multisensor systems. A consensus group approach which accounts for interdependence between group sensors is presented. The approach is based on an information theory formulation. An algorithm for computing a weighting scheme for the group sensors is developed. The algorithm rewards each sensor for the quality of its measurements and penalizes it for its interdependence with other sensors in the group.

The authors presented an approach that can be used to integrate multisensory data. The approach treats the sensors of a multisensor system as a group of experts cooperating with each other to resolve environment ambiguities. The approach is based on a recursive team utility function that is capable of bringing the team members to a consensus. This approach is simple and computationally tractable. The approach can be applied to integrate observation in object identification, a prominent area in which multisensor systems have the potential to play an important role.>

Resource allocation in a flexible manufacturing system (FMS), such as assignment of resources to tasks, can be solved efficiently using graph matching methods. The FMS is modeled as a graph: the vertices are resources and tasks, and the edges ar relationships between the resources and tasks. Resources are allocated by finding a match in the graph. The general matching problem is reviewed as an optimization problem; the more specific maximum cardinality 1-matching problem is discussed in detail. A direct parallelization of an algorithm is described which may be regarded as an archetype for implementation on a distributed network of computers in a large FMS. In such an FMS, planning activities such as resource allocation are decentralized, allowing for quicker and more detailed planning.>

Feature selection is an important phase in most pattern recognition problems, especially when the space of the extracted features is very large. Feature selection methods attempt to reduce the feature space to satisfy certain objectives. We propose the concept of defining a performance potential as a measure of the effectiveness of the set of selected features. This paper begins by outlining a ranking scheme for features based on a feature’s calculated “performance potential”. The performance potential is made up of a number of performance measures: extraction time, memory requirements, variance, covariance and classification success. An adaptive scheme is proposed to process a number of initial features and arrive at the “best” subset based on their performance potential. The approach is applied to a texture analysis problem. The results of the testing of the approach point to conclusions concerning its effectiveness.

This paper proposes a search-effective strategy for multi-class texture classification. The textures are classified according to the nearest neighbor rule based on our recently developed texture metric. We will show that it is possible to significantly reduce the amount of computation from an exhaustive search scheme. For this purpose, a distance-preserving vector space representation of the texture database is constructed. The representation facilitates the selection of a subset of class prototypes which constitute the reduced search space. In addition, the prototypes are organized into a hierarchy to further economize the search for the nearest class. This methodology is demonstrated by experiments on 720 texture samples belonging to eight classes. On average, a reduction of close to 70% is achieved.

In the field of electronic fault detection much of the evaluation is performed by a human operator using a visual inspection system. One such system displays the current versus voltage (I-V) characteristics, referred to as the impedance trace, of the electronic component under test. The shape of the impedance trace indicates the integrity of the component. The subjectivity and poor repeatability of the visual inspection have been identified as major deficiencies in the existing systems. A microprocessor-based system has been proposed to store and provide a quantitative classification of the impedance traces. The primary objective of this system is to provide an efficient automated system for classifying the impedance traces of different electronic components. Thus three trace representations and two classification schemes are proposed, based upon techniques developed in the field of pattern recognition. An experiment contrasts the different classification approaches. A Fourier series representation, combining both magnitude and phase features, was determined to provide the best overall performance when used in conjunction with a developed distance measure.