Xiaoyan Mu

A Weighted Voting Model of Associative Memory

This paper presents an analysis of a random access memory (RAM)-based associative memory which uses a weighted voting scheme for information retrieval. This weighted voting memory can operate in heteroassociative or autoassociative mode, can store both real-valued and binary-valued patterns and, unlike memory models, is equipped with a rejection mechanism. A theoretical analysis of the performance of the weighted voting memory is given for the case of binary and random memory sets. Performance measures are derived as a function of the model parameters: pattern size, window size, and number of patterns in the memory set. It is shown that the weighted voting model has large capacity and error correction. The results show that the weighted voting model can successfully achieve high-detection and -identification rates and, simultaneously, low-false-acceptance rates

An RCE-based Associative Memory with Application to Human Face Recognition

Many models of neural network-based associative memory have been proposed and studied. However, most of these models do not have a rejection mechanism and hence are not practical for many real-world associative memory problems. For example, in human face recognition, we are given a database of face images and the identity of each image. Given an input image, the task is to associate when appropriate the image with the corresponding name of the person in the database. However, the input image may be that of a stranger. In this case, the system should reject the input. In this paper, we propose a practical associative memory model that has a rejection mechanism. The structure of the model is based on the restricted Coulomb energy (RCE) network. The capacity of the proposed memory is desibed by two measures: the ability of the system to correctly identify known individuals, and the ability of the system to reject individuals who are not in the database. Experimental results are given which show how the performance of the system varies as the size of the database increases up to 1000 individuals.

Combining Gabor features: summing vs. voting in human face recognition

Gabor wavelet-based feature extraction has been emerging as one of the most promising ways to represent human face image data. In this paper, we examine the performance of two types of classifiers that can be used with Gabor features. In the first classifier, the distance between two images is computed by summing the local distances among all the nodes. In the second classifier, a voting strategy is used In addition, we examine two types of shift optimization procedures. The first is the standard elastic graph matching algorithm, and the second is a constrained version of the algorithm. Experimental results indicate that the voting-based classifier with constrained elastic graph matching gives improved results.

Training algorithms for robust face recognition using a template-matching approach

This paper describes a complete face recognition system. The system uses a template matching approach along with a training algorithm for tuning the performance of the system to solve two types of problems simultaneously: 1) correct classification experiments which correctly recognize and identify individuals who are in the database; and 2) false positive experiments which reject individuals who are not part of the database. Experimental results are given which indicate that this training method is capable of consistently producing high correct classification rates and low false positive rates.

Local voting networks for human face recognition

We investigate a template matching-based classifier system which uses local distance computations and a voting scheme. The proposed system has a mechanism to reject unknown patterns, and provides invariance to small amounts of translation. Experimental results are presented for 3 different types of face recognition problems: classification experiments where we measure the ability of the system to identify known individuals; false positive experiments where we measure the ability of the system to reject images of unknown individuals; and temporal experiments, where we measure the ability of the system to recognize images of known individuals taken over a period of time (6 months). The results show that the proposed system performs well on all 3 of these problems.

Weighted voting-based ensemble classifiers with application to human face recognition and voice recognition

A recent trend in the field of pattern recognition has been the use of ensemble classifiers. If combined properly, the ensemble can achieve a higher identification rate than any individual classifier. Plurality voting is one of the most commonly used combination strategies. The performance of plurality voting can be improved if the decisions of different classifiers are weighted properly. In this paper, we both theoretically and experimentally analyze the performance of a weighted plurality voting combination strategy to combine the decisions of multiple classifiers. Theoretical expressions characterizing the performance of the weighted voting model are derived and the method is applied to the problem of human face recognition and voice recognition. The results show the advantage of employing weighted-voting-based ensemble classifiers in achieving high identification rates.

Analysis of a plurality voting-based combination of classifiers

In various studies, it has been demonstrated that combining the decisions of multiple classifiers can lead to better recognition results. Plurality voting is one of the most widely used combination strategies. In this paper, we both theoretically and experimentally analyze the performance of a plurality voting-based ensemble classifier. Theoretical expressions for system performance are derived as a function of the model parameters: N (number of classifiers), m (number of classes), and p (probability that a single classifier is correct). Experimental results on the problem of human face recognition show that the voting strategy can successfully achieve high detection and identification rates, and, simultaneously, low false acceptance rates.

A Weighted Voting and Sequential Combination of Classifiers Scheme for Human Face Recognition

In this paper, we examine the performance of a weighted voting classification strategy for human face recognition. Here, local template matching is used, but instead of summing the local distance measures, a weighted voting scheme based on rank information is used to combine the results of the local classifiers. This strategy can be used with any suitable features; for example, simple pixel features, or Gabor features, etc. If multiple features are available, we show how a sequential combination strategy can be devised to efficiently and reliably compute the final classifier output. Test results are presented for the problem of human face recognition on a large database of faces.

Analysis of a Plurality Voting-based Combination of Classifiers

In various studies, it has been demonstrated that combining the decisions of multiple classifiers can lead to better recognition result. Plurality voting is one of the most widely used combination strategies. In this paper, we both theoretically and experimentally analyze the performance of a plurality voting based ensemble classifier. Theoretical expressions for system performance are derived as a function of the model parameters: N (number of classifiers), m (number of classes), and p (probability that a single classifier is correct). Experimental results on the human face recognition problem show that the voting strategy can successfully achieve high detection and identification rates, and, simultaneously, low false acceptance rates.

Combining local similarity measures: summing, voting, and weighted voting

Recent research on human face recognition has shown that local features have advantages over global features because local features are more robust to some changes of facial expression, as well as shift, rotation and tilt. In this paper, we experimentally investigate the commonly used summing strategy as well as the voting method in combining the local distances/similarity into the final decision. We proposed and analyzed a new classification method based on weighted voting that allows for each local window to cast not just a single vote, but a set of weighted votes. Experimental results are given on two large face databases: the CNNL and FERET databases. The results show that the weighted voting strategy outperforms simple voting, and the commonly used method of summing local distances.