Luis Rueda

Advances in Image and Video Technology

This book constitutes the refereed proceedings of the Second Pacific Rim Symposium on Image and Video Technology, PSIVT 2007, held in Santiago, Chile, in December 2007. The 75 revised full papers presented together with four keynote lectures were carefully reviewed and selected from 155 submissions. The symposium features ongoing research including all aspects of video and multimedia, both technical and artistic perspectives and both theoretical and practical issues.

Stochastic learning-based weak estimation of multinomial random variables and its applications to pattern recognition in non-stationary environments

In this paper, we formally present a novel estimation method, referred to as the Stochastic Learning Weak Estimator (SLWE), which yields the estimate of the parameters of a binomial distribution, where the convergence of the estimate is weak, i.e. with regard to the first and second moments. The estimation is based on the principles of stochastic learning. The mean of the final estimate is independent of the schemes learning coefficient, @l, and both the variance of the final distribution and the speed decrease with @l. Similar results are true for the multinomial case, except that the equations transform from being of a scalar type to be of a vector type. Amazingly enough, the speed of the latter only depends on the same parameter, @l, which turns out to be the only non-unity eigenvalue of the underlying stochastic matrix that determines the time-dependence of the estimates. An empirical analysis on synthetic data shows the advantages of the scheme for non-stationary distributions. The paper also briefly reports (without detailed explanation) conclusive results that demonstrate the superiority of SLWE in pattern-recognition-based data compression, where the underlying data distribution is non-stationary. Finally, and more importantly, the paper includes the results of two pattern recognition exercises, the first of which involves artificial data, and the second which involves the recognition of the types of data that are present in news reports of the Canadian Broadcasting Corporation (CBC). The superiority of the SLWE in both these cases is demonstrated.

Linear dimensionality reduction by maximizing the Chernoff distance in the transformed space

Linear dimensionality reduction (LDR) techniques are quite important in pattern recognition due to their linear time complexity and simplicity. In this paper, we present a novel LDR technique which, though linear, aims to maximize the Chernoff distance in the transformed space; thus, augmenting the class separability in such a space. We present the corresponding criterion, which is maximized via a gradient-based algorithm, and provide convergence and initialization proofs. We have performed a comprehensive performance analysis of our method combined with two well-known classifiers, linear and quadratic, on synthetic and real-life data, and compared it with other LDR techniques. The results on synthetic and standard real-life data sets show that the proposed criterion outperforms the latter when combined with both linear and quadratic classifiers.

Spot Detection and Image Segmentation in DNA Microarray Data

Following the invention of microarrays in 1994, the development and applications of this technology have grown exponentially. The numerous applications of microarray technology include clinical diagnosis and treatment, drug design and discovery, tumour detection, and environmental health research. One of the key issues in the experimental approaches utilising microarrays is to extract quantitative information from the spots, which represent genes in a given experiment. For this process, the initial stages are important and they influence future steps in the analysis. Identifying the spots and separating the background from the foreground is a fundamental problem in DNA microarray data analysis. In this review, we present an overview of state-of-the-art methods for microarray image segmentation. We discuss the foundations of the circle-shaped approach, adaptive shape segmentation, histogram-based methods and the recently introduced clustering-based techniques. We analytically show that clustering-based techniques are equivalent to the one-dimensional, standard k-means clustering algorithm that utilises the Euclidean distance.

A fully automatic gridding method for cDNA microarray images

BackgroundProcessing cDNA microarray images is a crucial step in gene expression analysis, since any errors in early stages affect subsequent steps, leading to possibly erroneous biological conclusions. When processing the underlying images, accurately separating the sub-grids and spots is extremely important for subsequent steps that include segmentation, quantification, normalization and clustering.ResultsWe propose a parameterless and fully automatic approach that first detects the sub-grids given the entire microarray image, and then detects the locations of the spots in each sub-grid. The approach, first, detects and corrects rotations in the images by applying an affine transformation, followed by a polynomial-time optimal multi-level thresholding algorithm used to find the positions of the sub-grids in the image and the positions of the spots in each sub-grid. Additionally, a new validity index is proposed in order to find the correct number of sub-grids in the image, and the correct number of spots in each sub-grid. Moreover, a refinement procedure is used to correct possible misalignments and increase the accuracy of the method.ConclusionsExtensive experiments on real-life microarray images and a comparison to other methods show that the proposed method performs these tasks fully automatically and with a very high degree of accuracy. Moreover, unlike previous methods, the proposed approach can be used in various type of microarray images with different resolutions and spot sizes and does not need any parameter to be adjusted.

A Hill-Climbing Approach for Automatic Gridding of cDNA Microarray Images

Image and statistical analysis are two important stages of cDNA microarrays. Of these, gridding is necessary to accurately identify the location of each spot while extracting spot intensities from the microarray images and automating this procedure permits high-throughput analysis. Due to the deficiencies of the equipment used to print the arrays, rotations, misalignments, high contamination with noise and artifacts, and the enormous amount of data generated, solving the gridding problem by means of an automatic system is not trivial. Existing techniques to solve the automatic grid segmentation problem cover only limited aspects of this challenging problem and require the user to specify the size of the spots, the number of rows and columns in the grid, and boundary conditions. In this paper, a hill-climbing automatic gridding and spot quantification technique is proposed which takes a microarray image (or a subgrid) as input and makes no assumptions about the size of the spots, rows, and columns in the grid. The proposed method is based on a hill-climbing approach that utilizes different objective functions. The method has been found to effectively detect the grids on microarray images drawn from databases from GEO and the Stanford genomic laboratories

An Efficient Algorithm for Optimal Multilevel Thresholding of Irregularly Sampled Histograms

Optimal multilevel thresholding is a quite importantproblem in image segmentation and pattern recognition. Althoughefficient algorithms have been proposed recently, they do notaddress the issue of irregularly sampled histograms. Apolynomial-time algorithm for multilevel thresholding ofirregularly sampled histograms is proposed. The algorithm ispolynomial not just on the number of bins of the histogram,n , but also on the number of thresholds, k , i.e.it runs in θ (kn 2). The proposedalgorithm is general enough for a wide range of thresholding andclustering criteria, and has the capability of dealing withirregularly sampled histograms. This implies important consequenceson pattern recognition, since optimal clustering in theone-dimensional space can be obtained in polynomial time.Experiments on synthetic and real-life histograms show that fortypical cases, the proposed algorithm can find the optimalthresholds in a fraction of a second.

Prediction of biological protein–protein interactions using atom-type and amino acid properties

Identification and analysis of types of biological protein–protein interactions and their interfaces to predict obligate and non‐obligate complexes is a problem that has drawn the attention of the research community in the past few years. In this paper, we propose a prediction approach to predict these two types of complexes. We use desolvation energies – amino acid and atom type – of the residues present in the interface. The prediction is performed via two state‐of‐the‐art classification techniques, namely linear dimensionality reduction (LDR) and support vector machines (SVM). The results on a newly compiled data set, namely BPPI, which is a joint and modified version of two well‐known data sets consisting of 213 obligate and 303 non‐obligate complexes, show that the best prediction is achieved with SVM (76.94% accuracy) when using desolvation energies of atom‐type features. Also, the proposed approach outperforms the previous solvent accessible area‐based approaches using SVM (75% accuracy) and LDR (73.06% accuracy). Moreover, a visual analysis of desolvation energies in obligate and non‐obligate complexes shows that a few atom‐type pairs are good descriptors for these types of complexes.

A formal analysis of why heuristic functions work

Many optimization problems in computer science have been proven to be NP-hard, and it is unlikely that polynomial-time algorithms that solve these problems exist unless P = NP. Alternatively, they are solved using heuristics algorithms, which provide a sub-optimal solution that, hopefully, is arbitrarily close to the optimal. Such problems are found in a wide range of applications, including artificial intelligence, game theory, graph partitioning, database query optimization, etc. Consider a heuristic algorithm, A. Suppose that A could invoke one of two possible heuristic functions. The question of determining which heuristic function is superior, has typically demanded a yes/no answer--one which is often substantiated by empirical evidence. In this paper, by using Pattern Classification Techniques (PCT), we propose a formal, rigorous theoretical model that provides a stochastic answer to this problem. We prove that given a heuristic algorithm, A, that could utilize either of two heuristic functions H1 or H2 used to find the solution to a particular problem, if the accuracy of evaluating the cost of the optimal solution by using H1 is greater than the accuracy of evaluating the cost using H2, then H1 has a higher probability than H2 of leading to the optimal solution. This unproven conjecture has been the basis for designing numerous algorithms such as the A* algorithm, and its variants. Apart from formally proving the result, we also address the corresponding database query optimization problem that has been open for at least two decades. To validate our proofs, we report empirical results on database query optimization techniques involving a few well-known histogram estimation methods.

The role of electrostatic energy in prediction of obligate protein-protein interactions.

BackgroundPrediction and analysis of protein-protein interactions (PPI) and specifically types of PPIs is an important problem in life science research because of the fundamental roles of PPIs in many biological processes in living cells. In addition, electrostatic interactions are important in understanding inter-molecular interactions, since they are long-range, and because of their influence in charged molecules. This is the main motivation for using electrostatic energy for prediction of PPI types.ResultsWe propose a prediction model to analyze protein interaction types, namely obligate and non-obligate, using electrostatic energy values as properties. The prediction approach uses electrostatic energy values for pairs of atoms and amino acids present in interfaces where the interaction occurs. The main features of the complexes are found and then the prediction is performed via several state-of-the-art classification techniques, including linear dimensionality reduction (LDR), support vector machine (SVM), naive Bayes (NB) and k-nearest neighbor (k-NN). For an in-depth analysis of classification results, some other experiments were performed by varying the distance cutoffs between atom pairs of interacting chains, ranging from 5Å to 13Å. Moreover, several feature selection algorithms including gain ratio (GR), information gain (IG), chi-square (Chi2) and minimum redundancy maximum relevance (mRMR) are applied on the available datasets to obtain more discriminative pairs of atom types and amino acid types as features for prediction.ConclusionsOur results on two well-known datasets of obligate and non-obligate complexes confirm that electrostatic energy is an important property to predict obligate and non-obligate protein interaction types on the basis of all the experimental results, achieving accuracies of over 98%. Furthermore, a comparison performed by changing the distance cutoff demonstrates that the best values for prediction of PPI types using electrostatic energy range from 9Å to 12Å, which show that electrostatic interactions are long-range and cover a broader area in the interface. In addition, the results on using feature selection before prediction confirm that (a) a few pairs of atoms and amino acids are appropriate for prediction, and (b) prediction performance can be improved by eliminating irrelevant and noisy features and selecting the most discriminative ones.