Combining Correlated Data from Multiple Classifiers

DOI: 10.1007/978-3-642-03625-5_11


Real time data collected from sensors and then having been processed by multiple classifiers are correlated due to common,
inherent noise resulting from the sensor. The data is collected at different ranges and may have different statistical distributions
and characteristics. Measurements from different classifiers are fused together to obtain more information about the phenomenon
or environment under observation. Since the classifier fusion is attempting to improve the data through processing only, this
chapter focuses on object refinement or fusion level 1 within the Joint Directors of Laboratories (JDL) data fusion process
model and taxonomy of algorithms. Before fusion, the correlated data is normalized using a variety of normalization techniques
such as z-score and min-max normalization. Further, by dynamically weighting and combining the measurements from these classifiers,
performance is improved. Weights are found using a Particle Swarm Optimization search algorithm and are a function of required
accuracy and degree of correlation. Results are presented for (a) synthetic data generated using conditional multivariate
normal distribution with different covariance matrices and (b) NIST BSSR dataset. For correlated classifiers, the performance
of the particle swarm optimization technique outperforms the traditional data level fusion i.e., z-normalization and min-max.

7 Reads
  • [Show abstract] [Hide abstract]
    ABSTRACT: The extension of classical detection theory, based on the theory of statistical hypothesis testing, to the case of distributed sensors is discussed. The development is based on the formulation of a decentralized or team hypothesis testing problem. Theoretical results concerning the form of the optimal decision rule, examples, application to data fusion, and open problems are presented.
    Decision and Control including the Symposium on Adaptive Processes, 1980 19th IEEE Conference on; 07/1981
  • Conference Paper: On Combining Classifiers
    [Show abstract] [Hide abstract]
    ABSTRACT: We develop a common theoretical framework for combining classifiers which use distinct pattern representations and show that many existing schemes can be considered as special cases of compound classification where all the pattern representations are used jointly to make a decision. An experimental comparison of various classifier combination schemes demonstrates that the combination rule developed under the most restrictive assumptions-the sum rule-and its derivatives consistently outperform other classifier combinations schemes. A sensitivity analysis of the various schemes to estimation errors is carried out to show that this finding can be justified theoretically
    Pattern Recognition, 1996., Proceedings of the 13th International Conference on; 09/1996
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: This paper presents an evolutionary approach to the sensor management of a biometric security system that improves robustness. Multiple biometrics are fused at the decision level to support a system that can meet more challenging and varying accuracy requirements as well as address user needs such as ease of use and universality better than a single biometric system or static multimodal biometric system. The decision fusion rules are adapted to meet the varying system needs by particle swarm optimization, which is an evolutionary algorithm. This paper focuses on the details of this new sensor management algorithm and demonstrates its effectiveness. The evolutionary nature of adaptive, multimodal biometric management (AMBM) allows it to react in pseudoreal time to changing security needs as well as user needs. Error weights are modified to reflect the security and user needs of the system. The AMBM algorithm selects the fusion rule and sensor operating points to optimize system performance in terms of accuracy.
    IEEE Transactions on Systems Man and Cybernetics Part C (Applications and Reviews) 09/2005; 35(3-35):344 - 356. DOI:10.1109/TSMCC.2005.848191 · 2.17 Impact Factor
Show more