International Journal of Pattern Recognition and Artificial Intelligence

Irrespective of the way computational intelligence (CI) is defined, its components should have the following characteristics: considerable potential in solving real world problems, ability to learn from experience, capability of self-organizing, and ability of adapting in response to dynamically changing conditions and constraints. To summarize, it should display aspects of intelligent behavior as observed in humans.

In view of these, we assume that the major ingredients of a computational intelligence system are artificial neural networks, fuzzy sets, rough sets, and evolutionary computation. Some other components that may be parts of computational intelligence (CI) systems are artificial life and immuno computing. It is a synergistic combination of all these components.

This paper reports an efficient technique of evolving Cellular Automata (CA) as an associative memory model. The evolved CA termed as GMACA (Generalized Multiple Attractor Cellular Automata), acts as a powerful pattern recognizer. Detailed analysis of GMACA rules establishes the fact that the rule subspace of the pattern recognizing CA lies at the edge of chaos — believed to be capable of executing complex computation.

A self-organizing neural network model that computes the smallest circle (also called minimum spanning circle) enclosing a finite set of given points was proposed by Datta.³ In the article,³ the algorithm is stated and it is demonstrated by simulation that the center of the smallest circle can be achieved with a given level of accuracy. No rigorous proof was given in support of the simulation results. In this paper, we make a rigorous analysis of the model and mathematically prove that the model converges to the desired center of the minimum spanning circle. A suitable neural network architecture is also designed for parallel implementation of the proposed model. Time complexity of the algorithm is worked out under the proposed architecture. Extension of the proposed model to higher dimensions is discussed and demonstrated with some applications.

Different eigenspace-based approaches have been proposed for the recognition of faces. They differ mostly in the kind of projection method being used and in the similarity matching criterion employed. The aim of this paper is to present a comparative study between some of these different approaches. This study considers theoretical aspects as well as experiments performed using a face database with a few number of classes (Yale) and also with a large number of classes (FERET).

Bioinformatics has recently drawn a lot of attention to efficiently analyze biological genomic information with information technology, especially pattern recognition. In this paper, we attempt to explore extensive features and classifiers through a comparative study of the most promising feature selection methods and machine learning classifiers. The gene information from a patient's marrow expressed by DNA microarray, which is either the acute myeloid leukemia or acute lymphoblastic leukemia, is used to predict the cancer class. Pearson's and Spearman's correlation coefficients, Euclidean distance, cosine coefficient, information gain, mutual information and signal to noise ratio have been used for feature selection. Backpropagation neural network, self-organizing map, structure adaptive self-organizing map, support vector machine, inductive decision tree and k-nearest neighbor have been used for classification. Experimental results indicate that backpropagation neural network with Pearson's correlation coefficients produces the best result, 97.1% of recognition rate on the test data.

This paper proposes a novel approach to automatic recognition of handprinted Bangla (an Indian script) numerals. A modified Topology Adaptive Self-Organizing Neural Network is proposed to extract a vector skeleton from a binary numeral image. Simple heuristics are considered to prune artifacts, if any, in such a skeletal shape. Certain topological and structural features like loops, junctions, positions of terminal nodes, etc. are used along with a hierarchical tree classifier to classify handwritten numerals into smaller subgroups. Multilayer perceptron (MLP) networks are then employed to uniquely classify the numerals belonging to each subgroup. The system is trained using a sample data set of 1800 numerals and we have obtained 93.26% correct recognition rate and 1.71% rejection on a separate test set of another 7760 samples. In addition, a validation set consisting of 1440 samples has been used to determine the termination of the training algorithm of the MLP networks. The proposed scheme is sufficiently robust with respect to considerable object noise.

In 1982, Pawlak proposed the concept of rough sets with a practical purpose of representing indiscernibility of elements or objects in the presence of information systems. Even if it is easy to analyze, the rough set theory built on a partition induced by equivalence relation may not provide a realistic view of relationships between elements in real-world applications. Here, coverings of, or nonequivalence relations on, the universe can be considered to represent a more realistic model instead of a partition in which a generalized model of rough sets was proposed. In this paper, first a weak fuzzy similarity relation is introduced as a more realistic relation in representing the relationship between two elements of data in real-world applications. Fuzzy conditional probability relation is considered as a concrete example of the weak fuzzy similarity relation. Coverings of the universe is provided by fuzzy conditional probability relations. Generalized concepts of rough approximations and rough membership functions are proposed and defined based on coverings of the universe. Such generalization is considered as a kind of fuzzy rough set. A more generalized fuzzy rough set approximation of a given fuzzy set is proposed and discussed as an alternative to provide interval-value fuzzy sets. Their properties are examined.

Stereo matching is the central problem of stereovision paradigm. Area-based techniques provide the dense disparity maps and hence they are preferred for stereo correspondence. Normalized cross correlation (NCC), sum of squared differences (SSD) and sum of absolute differences (SAD) are the linear correlation measures generally used in the area-based techniques for stereo matching. In this paper, similarity measure for stereo matching based on fuzzy relations is used to establish the correspondence in the presence of intensity variations in stereo images. The strength of relationship of fuzzified data of two windows in the left image and the right image of stereo image pair is determined by considering the appropriate fuzzy aggregation operators. However, these measures fail to establish correspondence of the pixels in the stereo images in the presence of occluded pixels in the corresponding windows. Another stereo matching algorithm based on fuzzy relations of fuzzy data is used for stereo matching in such regions of images. This algorithm is based on weighted normalized cross correlation (WNCC) of the intensity data in the left and the right windows of stereo image pair. The properties of the similarity measures used in these algorithms are also discussed. Experiments with various real stereo images prove the superiority of these algorithms over normalized cross correlation (NCC) under nonideal conditions.

In this paper we present a voting scheme for fuzzy cluster algorithms. This voting method allows us to combine several runs of cluster algorithms resulting in a common partition. This helps us to tackle the problem of choosing the appropriate clustering method for a data set where we have no a priori information about it. We mathematically derive the algorithm from theoretical considerations. Experiments show that the voting algorithm finds structurally stable results. Several cluster validity indexes show the improvement of the voting result in comparison to simple fuzzy voting.

This paper proposes a fuzzy approach to speaker verification. For an input utterance and a claimed identity, most of the current methods compute a claimed speaker's score, which is the ratio of the claimed speaker's and the impostors' likelihood functions, and compare this score with a given threshold to accept or reject this speaker. Considering the speaker verification problem based on fuzzy set theory, the claimed speaker's score is viewed as the fuzzy membership function of the input utterance in the claimed speaker's fuzzy set of utterances. Fuzzy entropy and fuzzy c-means membership functions are proposed as fuzzy membership scores, which are the ratios of functions of the claimed speaker's and impostors' likelihood functions. A likelihood transformation is also considered to relate current likelihood and fuzzy membership scores. We also proposed fuzzy scores using membership functions similar to those produced by noise-clustering-based method. This noise clustering concept provides very effective modifications to several methods, which can overcome some of the problems of ratio-type scores and greatly reduce the false acceptance rate. Experiments were performed to evaluate proposed normalization methods for speaker verification using the YOHO corpus. Experiments demonstrate that fuzzy methods and their noise clustering versions outperform conventional methods.

This paper addresses the problem of speech segmentation and enhancement in the presence of noise. We first propose a new word boundary detection algorithm by using a neural fuzzy network (called ATF-based SONFIN algorithm) for identifying islands of word signals in fixed noise-level environment. We further propose a new RTF-based RSONFIN algorithm where the background noise level varies during the procedure of recording. The adaptive time-frequency (ATF) and refined time-frequency (RTF) parameters extend the TF parameter from single band to multiband spectrum analysis, and help to make the distinction of speech and noise signals clear. The ATF and RTF parameters can extract useful frequency information by adaptively choosing proper bands of the mel-scale frequency bank. Due to the self-learning ability of SONFIN and RSONFIN, the proposed algorithms avoid the need of empirically determining thresholds and ambiguous rules. The RTF-based RSONFIN algorithm can also find the variation of the background noise level and detect correct word boundaries in the condition of variable background noise level by processing the temporal relations. Our experimental results show that both in the fixed and variable noise-level environment, the algorithms that we proposed achieved higher recognition rate than several commonly used word boundary detection algorithms and reduced the recognition error rate due to endpoint detection.

This paper presents an evolvable version of a novel subsethood product fuzzy neural inference system (ESuPFuNIS). The original SuPFuNIS model²⁰ employs only fuzzy weights, and accepts both numeric and linguistic inputs. All numeric inputs are fuzzified using a feature specific fuzzifier. The model composes fuzzy signals from the input layer with fuzzy weights using a mutual subsethood measure. Rule nodes use a product aggregation operator. Outputs from the network are generated using volume defuzzification. Here we replace the original gradient descent learning procedure with a genetic optimization technique and report considerable improvements in classification accuracy and rule economy on three benchmark problems. Real-coded genetic algorithms (RGA's) have been employed to search for an optimal set of network parameters. We demonstrate the classification capabilities of the network on Ripley's synthetic two class data, Iris data and Forensic glass data. In all the problems considered, the GA based classifier performs better than its gradient descent counterpart in terms of classification accuracy as well as rule economy.