Advances in Intelligent Information Processing

The following sections are included:

• Foreword

• Preface

• Contents

We present a novel region-based level set algorithm. We first model the image histogram with a mixture of non-parametric probability density functions (PDFs), whose use we justify. The individual densities in the mixture are estimated using a recently proposed PDF estimation method which relies on a continuous representation of discrete signals. A Bayesian framework is then formulated in which likelihood probabilities are given by the non-parametric PDFs and prior probabilities are calculated using an inequality constrained least squares method. The segmentation solution is spatially regularised using a level sets framework. The log ratio of the posterior probabilities is used to drive the level set evolution. We also take into account the partial volume effect, which is important in medical image analysis. Results are presented on natural, as well as medical, two-dimensional images. Visual inspection of results on a range of images show the effectiveness of the proposed algorithm.

Patterns are being generated in nature through biological and chemical processes. In this paper we are proposing artificial pattern generation technique using curve evolution model. Given a closed curve in 2D space the curve is deformed under a set of constraints derived from biological and physical pattern generation models. In the proposed approach the reaction-diffusion and shape optimization models are used to derive the constraints for curve evolution. The evolution of curve is implemented in level set framework as the level set based curve evolution supports change in topology of the closed contour. The proposed model is used to generate number of patterns and also successfully tested for reconstructing partially occluded patterns.

Active contours are very widely used in computer vision problems. Their usage has a typical problem: bunching together of curve points. This becomes apparent especially when we use active contours for object tracking leading to instability in curve evolution. In this paper, we propose an additional tangential component to stabilise the evolution while at the same time ensuring that the curve shape is not changed. The proposed method is simple and the computational overhead is minimal, while the results are very good.

This paper describes an information theoretic approach for next best view planning in active state estimation, and its application to three computer vision tasks. In active state estimation, the state estimation process contains sensor actions which affect the state observation, and therefore the final state estimate. We use the information theoretic measure of mutual information to quantify the information content in this estimate. The optimal sensor actions are those that are expected to maximally increase the information content of the estimate. This action selection process is then applied to three seperate computer vision tasks: object recognition, object tracking and object reconstruction. Each task is formulated as an active state estimation problem. In these tasks, a given sensor action describes a camera position, or view. The information theoretic framework allows us to determine the next best view, i.e. the view that best supports the computer vision task. We show the benefits of next best view planning in several experiments, in which we compare the estimation error produced by planned views with the error produced by regularly sampled or unchanging views.

This action selection process is then applied to three seperate computer vision tasks: object recognition, object tracking and object reconstruction. Each task is formulated as an active state estimation problem. In these tasks, a given sensor action describes a camera position, or view. The information theoretic framework allows us to determine the next best view, i.e. the view that best supports the computer vision task.

We show the benefits of next best view planning in several experiments, in which we compare the estimation error produced by planned views with the error produced by regularly sampled or unchanging views.

In this work, we present a method for model-based recognition of 3d objects from a small number of 2d intensity images taken from nearby, but otherwise arbitrary viewpoints. Our method works by linearly combining images from two (or more) viewpoints of a 3d object to synthesise novel views of the object. The object is recognised in a target image by matching to such a synthesised, novel view. All that is required is the recovery of the linear combination parameters, and since we are working directly with pixel intensities, we suggest searching the parameter space using a global, evolutionary optimisation algorithm combined with a local search method in order efficiently to recover the optimal parameters and thus recognise the object in the scene. We have experimented with both synthetic data and real-image, public databases.

In this paper we present a framework which integrates object-model knowledge with the perceptual organization process. We demonstrate the advantages of the add-on grouping evidences as contributed by the object models for a more robust perceptual organization in the spatio-temporal domain. Our system performs detection of foreground objects along with recognition in video.

In this chapter, we present a novel algorithm for extracting features in a supervised manner. This algorithm is derived from the local non-negative matrix factorization (LNMF) algorithm, which is an extension of non-negative matrix factorization (NMF) method. We call this newly proposed algorithm Discriminant Non-negative Matrix Factorization (DNMF). Given an image database, all these three algorithms decompose the image database into basis images and their corresponding coefficients. This decomposition is computed differently for each method. The decomposition results are applied on facial images for the recognition of the six basic facial expressions. We found that our algorithm shows superior performance by achieving a higher recognition rate, when compared to NMF and LNMF. Moreover, we compared DNMF classification performance with principal component analysis (PCA), Fisher Linear Discriminant (FLD), independent component analysis (ICA) and Gabor approaches. We found they are outperformed by DNMF. The experiments were conducted for a varying number of basis images (subspaces). The recognition rate of the proposed algorithm seems to be more robust than that of other methods with respect to the choice of the number of basis images.

We propose an image duplicate detection method for identifying modified copies of the same image in a very large database. Modifications that we consider include rotation, scaling and cropping. A compact 12 dimensional descriptor based on Fourier Mellin Transform is introduced. The compactness of this descriptor allows efficient indexing over the entire database. Results are presented on a 10 million image database that demonstrates the effectiveness and the efficiency of this descriptor. In addition, we also propose extension to arbitrary shape representations and similar scene detection and preliminary results are also included.

In this article unsupervised context-sensitive techniques for change detection in multitemporal remote sensing images have been proposed. Two different architecture of Self-Organizing Feature Map Neural Network are used to discriminates the changed and unchanged regions of the difference image. In both the networks, the number of neurons in the input layer is equal to the number of features of the input patterns. Depending on the number of neurons in the output layer the network is categorized as one-dimensional or two-dimensional. In one-dimensional network, the number of neuron in output layer is two, one represents the changed class and the other represents the unchanged class. In two-dimensional network, each spatial position of the input image corresponds to a neuron in the output layer and the network is updated depending on some threshold value. When the network converges, status of output neurons depict a change detection map. Two different criteria are suggested to select suitable thresholds. Both the proposed change detection techniques are distribution free. Experimental results are carried out on two multispectral and multitemporal remote sensing images to assess the effectiveness of the proposed approaches.

In the last two decades, there have been significant advances in video coding. Since the early 1990s, a series of video coding standards have been developed to satisfy the growing requirements of video applications. Among these, H.264/MPEG-4 AVC is the most recent standard achieving high compression efficiency. The video compression research community has also continued working on new advances that go beyond traditional video coding architectures. In this paper, we give an overview of some recent advances in video coding and their potential applications. We discuss the most recent video coding standard and work in distributed coding, texture-based methods, scalable coding, and multi-view video coding.

Automatic Fingerprint Identification Systems (AFIS) have various applications in biometric authentication, forensic decision, and in many other areas. Most of the fingerprint matching techniques require extraction of ridge lines and/or minutiae for processing. Minutiae are the terminations and bifurcations of the ridge lines in a fingerprint image. Detection of ridge lines from a noisy gray-scale fingerprint image is a challenging task. In this work, a novel combinatorial approach is proposed for designing a hardware architecture aimed at classifying each pixel into one of the three classes (crest, valley, and slope), based on its gray-scale topographical relationship with its neighbors. A two-pass algorithm is developed for the pixel classification scheme. The ridge lines are then detected as formed by the thinned version of the crest pixels. The algorithm is robust, performs very well in the presence of noise, and has minimal dependence on thresholding. It has been tested on several benchmark fingerprint images and is observed to produce good results both in terms of quality of solutions and CPU time. A simple pipelined design is proposed for its hardware implementation. The design does not need any floating point operation, and hence, can be used to build a fast coprocessor for online fingerprint analysis.

In most pattern recognition algorithms, amino acids cannot be used directly as inputs since they are nonnumerical variables. They, therefore, need encoding prior to input. In this regard, bio-basis function maps a nonnumerical sequence space to a numerical feature space. It is designed using an amino acid mutation matrix. One of the important issues for the bio-basis function is how to select a minimum set of bio-bases with maximum information. In this chapter, we describe an algorithm, termed as rough-fuzzy c-medoids algorithm, to select most informative bio-bases. It comprises a judicious integration of the principles of rough sets, fuzzy sets, c-medoids algorithm, and amino acid mutation matrix. While the membership function of fuzzy sets enables efficient handling of overlapping partitions, the concept of lower and upper bounds of rough sets deals with uncertainty, vagueness, and incompleteness in class definition. The concept of crisp lower bound and fuzzy boundary of a class, introduced in rough-fuzzy c-medoids, enables efficient selection of a minimum set of most informative bio-bases. Some indices are introduced for evaluating quantitatively the quality of selected biobases. The effectiveness of the algorithm, along with a comparison with other algorithms, has been demonstrated on different types of protein data sets.

In this study, we discuss the concept of knowledge reuse in Computational Intelligence (CI) by showing how the CI constructs could be formed by reconciling data and a suite of existing models. As the models are reflective of the available knowledge being formed on a basis of existing data, their use is effectively a sound mechanism of knowledge reuse. We introduce a certain performance index whose minimization helps establish the most effective level of knowledge reuse (viz. a level of reliance on the existing models). Fuzzy rule-based systems are used as a comprehensive example which illustrates the algorithmic details of the proposed approach. Furthermore we show, by making use of the principle of justifiable granularity, how the parameters of the rules can be represented in the form of information granules, and triangular fuzzy numbers in particular.

The following sections are included:

• Index