Lattice computing (LC) meta-representation for pattern classification

Abstract

This paper compares two alternative feature data meta-representations using Intervals' Numbers (INs) in the context of the Minimum Distance Classifier (MDC) model. The first IN meta-representation employs one IN per feature vector, whereas the second IN meta-representation employs one IN per feature per class. Comparative classification experiments with the standard minimum distance classifier (MDC) on two benchmark classification problems, regarding face/facial expression recognition, demonstrate the superiority of the aforementioned second IN meta-representation. This superiority is attributed to an IN's capacity to represent discriminative, all-order data statistics in a population of features.

Learning Distributions of Image Features by Interactive Fuzzy Lattice Reasoning in Pattern Recognition Applications

“…In particular, we induced orthogonal moments as well as other features due to their practical effectiveness [16], [22], [26]. Then, a distribution of features is "meta-represented" by an IN [24] induced by algorithm CALCIN [13]. A recent work [27] has demonstrated specific advantages for an IN meta-representation including a significant dimensionality reduction as well as a superior pattern recognition performance.…”

“…This paper retains a basic Feature Extraction employed elsewhere [16], [22], [24], [26]; that is, a population of numerical features is induced from an image to be learned/recognized. In particular, we induced orthogonal moments as well as other features due to their practical effectiveness [16], [22], [26].…”

“…More specifically, we computed four different families of orthogonal moments including Zernike (ZMs), Gaussian-Hermite (GHMs), Tchebichef (TMs) and dual Hahn (dHMs) moments [23]; the order of each moment family was selected such that a 16-dimensional vector was produced. Another two types of features, popular in face recognition applications, were extracted, namely the Local Binary Pattern (LBP) and Histogram of Oriented Gradient (HOG) [24]. Regarding LBP, uniform patterns of ( , ) ( , ) R N 1 8 = regions were computed.…”

“…We employed ten traditional classifiers including three versions of the Minimum Distance Classifier (MDC) corresponding to the Chi Square ( ), 2 | Euclidean and Manhattan distances, respectively [26] -An MDC classif ier here engaged "mean feature vectors" [24]. In addition, we employed a kNN (k =1), a Naïve-Bayes, an RBF ELM, a three-layer feedforward backpropagation Neural Network, and three types of Support Vector C c of tunable parameters computed optimally by the grid search method [8].…”

“…This paper builds on recently published work [16], [22], [24], [26] regarding human face recognition based on INs induced from orthogonal moments features. Differences with this work are summarized next.…”