IMAGE CLASSIFICATION USING LOGICAL ANALYSIS OF DATA

Abstract

Convolutional Neural Networks (CNNs) dominate various computer vision tasks since Alex Krizhevsky showed that they can be trained effectively and reduced the top-5 error from 26.2 % to 15.3 % on the ImageNet large scale visual recognition challenge. Many aspects of CNNs are examined in various publications, however, CNNs come with a set of disadvantages and limitations. CNNs or any deep learning models have no interpretability by humans. The reasoning behind a prediction in a CNN can never be understood and that is a problem when building reliable and robust AI solutions. Another issue with the CNNs is that they require huge amounts of data to work well. Feature extraction from images has been a popular technique in many computer vision problems. This report proposes a solution based on feature extraction from techniques like SIFT, SURF, HOG, etc. and using representation learning for learning different features from training images. Logical Analysis of Data is then used to classify images from one class label to another to solve the interpretability problem.