Event Image Classification using Deep Learning
S. Regina Lourdhu Suganthi, M. Hanumanthappa, S. Kavitha · 2018
Efficient Image data storage and tagged Image Archive are quintessential in organizations due to the growing volume of images that are captured during various events that are conducted in organizations. The multiplicity of information and complexity of image content is proven to be challenging for the computer to understand and interpret an image. To retrieve an image from an image collection, image content should be automatically understood and interpreted by the machine. Automatic classification and annotation schemes using machine learning techniques are generally rigid. A scene in an image is a composition of multiple objects. In such semantic scenario, the feature space will not be mutually exclusive and thus give rise to classification error. Deep learning is an emerging machine learning technique which could be applied for various classification and prediction problems with high accuracy. In this work large event image data set is used to train the classifier for binary classification as indoor or outdoor event image. Image augmentation is used to generate multiple images to reduce bias and improve accuracy. To reduce the False Positive Rate an appropriate model with convolution neural network having multiple hidden layers of dynamic parameters are used. The model has been evaluated using the measures namely, Precision, Recall, and F-Score resulting in 80 to 85 percent accuracy on the test data.