Open AccessCategorized Image Classification uing CNN Features with ECOC Framework
Author(s) -
Shameem Fatima,
M Seshashayee
Publication year - 2019
Publication title -
international journal of recent technology and engineering
Language(s) - English
Resource type - Journals
ISSN - 2277-3878
DOI - 10.35940/ijrte.a1937.078219
Subject(s) - multiclass classification , artificial intelligence , pattern recognition (psychology) , computer science , support vector machine , classifier (uml) , contextual image classification , convolutional neural network , feature extraction , image (mathematics) , machine learning
Image Classification technique is used to classifyimages into categories. In this study, an application is presentedto examine category based image classification by combiningSupport Vector Machine with error correcting output codes(ECOC) framework. The ResNet50 used as Networkarchitecture, our image dataset include caltech101 images from9 categories (classes) which builds our classification task amulticlass problem. ECOC is a commonly used framework tomodel multiclass classification problem. We presentone-verses-all coding design of ECOC and apply to SVMclassifier. A pre-trained CNN (convolution neural network) isused for extracting image feature and as a classifier MulticlassSupport Vector Machine is used. The extracted features are thenpassed for classification via ECOC approach. The finalclassification result predicts the class labels. The application isimplemented in Matlab using pre-trained CNN. The predictionaccuracy of each category is evaluated and presented. Theexperimental result shows an accuracy of 97.6%. Furtherexperiments are carried out on different dataset which showedthat best accuracy is achieved using CNN with ECOC formulticlass problem.