A hybrid Deep Boltzmann Functional Link Network for classification problems
MetadataShow full item record
© 2016 IEEE.This paper proposes a hybrid deep learning algorithm, namely, the Deep Boltzmann Functional Link Network (DBFLN) for classification problems. A Deep Boltzmann Machine (DBM) with two layers of Restricted Boltzmann Machine is the generative model that is used to generate stochastic features and input weights for the discriminative model. A discriminative Functional Link Network (FLN) uses these features to approximate the nonlinear relationship between a set of features and their classes. FLN has three layers, namely, the input layer, the enhancement layer and the output layer. In a DBFLN, the features generated at the two hidden layers of the DBM act as the input features and the enhancement layer responses of the FLN. The output weights of the FLN are then estimated as a solution to a linear programming problem through pseudo-inverse. We first evaluate the performance of the DBFLN on three benchmark multi-category classification problems from the UCI machine learning repository, namely, the image segmentation problem, the vehicle classification problem and the glass identification problem. Performance study results on the benchmark classification problems show that DBFLN is an efficient classifier. We then use the DBFLN to classify the images in the TID2013 data set, based on their depth of distortions. The TID2013 data set comprises of 25 images, each with 5 levels of 24 distortion types. In all, the data set has 3000 images, which can be classified based on the depth of distortion. Thus, the IQA classification problem is defined as classifying the distorted images into one of the 5 classes (depending on the depth of distortion) using human visual image metrics as the input features. The performance of the DBFLN in classifying the image quality is compared with those of Support Vector Machines, Extreme Learning Machines, Random Vector Functional Link Network, and Deep Belief Network. Performance studies show the superior classification ability of the DBFLN.
Showing items related by title, author, creator and subject.
Hayat, M.; Bennamoun, M.; An, Senjian (2015)© 2015 IEEE. Image set classification finds its applications in a number of real-life scenarios such as classification from surveillance videos, multi-view camera networks and personal albums. Compared with single image ...
Estimating size fraction categories of coal particles on conveyor belts using image texture modeling methodsJemwa, G.; Aldrich, Chris (2012)Motivation: Physical properties of coal such as particle size distribution have a large influence on the stability and operational behavior of fluidized bed reactors and metallurgical furnaces. In particular, the presence ...
Hayat, M.; Bennamoun, M.; An, Senjian (2014)© 2014 IEEE. We propose a deep learning framework for image set classification with application to face recognition. An Adaptive Deep Network Template (ADNT) is defined whose parameters are initialized by performing ...