Date on Master's Thesis/Doctoral Dissertation


Document Type

Doctoral Dissertation

Degree Name

Ph. D.


Computer Engineering and Computer Science

Committee Chair

Frigui, Hichem

Author's Keywords

Information fusion; Clustering; Local fusion; Data mining; Landmine detection; Pattern recognition


Pattern perception--Data processing; Pattern recognition systems; Land mines--Detection; Data mining


For complex detection and classification problems, involving data with large intra-class variations and noisy inputs, no single source of information can provide a satisfactory solution. As a result, combination of multiple classifiers is playing an increasing role in solving these complex pattern recognition problems, and has proven to be a viable alternative to using a single classifier. Over the past few years, a variety of schemes have been proposed for combining multiple classifiers. Most of these were global as they assign a degree of worthiness to each classifier, that is averaged over the entire training data. This may not be the optimal way to combine the different experts since the behavior of each one may not be uniform over the different regions of the feature space. To overcome this issue, few local methods have been proposed in the last few years. Local fusion methods aim to adapt the classifiers' worthiness to different regions of the feature space. First, they partition the input samples. Then, they identify the best classifier for each partition and designate it as the expert for that partition. Unfortunately, current local methods are either computationally expensive and/or perform these two tasks independently of each other. However, feature space partition and algorithm selection are not independent and their optimization should be simultaneous. In this dissertation, we introduce a new local fusion approach, called Context Extraction for Local Fusion (CELF). CELF was designed to adapt the fusion to different regions of the feature space. It takes advantage of the strength of the different experts and overcome their limitations. First, we describe the baseline CELF algorithm. We formulate a novel objective function that combines context identification and multi-algorithm fusion criteria into a joint objective function. The context identification component thrives to partition the input feature space into different clusters (called contexts), while the fusion component thrives to learn the optimal fusion parameters within each cluster. Second, we propose several variations of CELF to deal with different applications scenario. In particular, we propose an extension that includes a feature discrimination component (CELF-FD). This version is advantageous when dealing with high dimensional feature spaces and/or when the number of features extracted by the individual algorithms varies significantly. CELF-CA is another extension of CELF that adds a regularization term to the objective function to introduce competition among the clusters and to find the optimal number of clusters in an unsupervised way. CELF-CA starts by partitioning the data into a large number of small clusters. As the algorithm progresses, adjacent clusters compete for data points, and clusters that lose the competition gradually become depleted and vanish. Third, we propose CELF-M that generalizes CELF to support multiple classes data sets. The baseline CELF and its extensions were formulated to use linear aggregation to combine the output of the different algorithms within each context. For some applications, this can be too restrictive and non-linear fusion may be needed. To address this potential drawback, we propose two other variations of CELF that use non-linear aggregation. The first one is based on Neural Networks (CELF-NN) and the second one is based on Fuzzy Integrals (CELF-FI). The latter one has the desirable property of assigning weights to subsets of classifiers to take into account the interaction between them. To test a new signature using CELF (or its variants), each algorithm would extract its set of features and assigns a confidence value. Then, the features are used to identify the best context, and the fusion parameters of this context are used to fuse the individual confidence values. For each variation of CELF, we formulate an objective function, derive the necessary conditions to optimize it, and construct an iterative algorithm. Then we use examples to illustrate the behavior of the algorithm, compare it to global fusion, and highlight its advantages. We apply our proposed fusion methods to the problem of landmine detection. We use data collected using Ground Penetration Radar (GPR) and Wideband Electro -Magnetic Induction (WEMI) sensors. We show that CELF (and its variants) can identify meaningful and coherent contexts (e.g. mines of same type, mines buried at the same site, etc.) and that different expert algorithms can be identified for the different contexts. In addition to the land mine detection application, we apply our approaches to semantic video indexing, image database categorization, and phoneme recognition. In all applications, we compare the performance of CELF with standard fusion methods, and show that our approach outperforms all these methods.