Multi Instance Multi Label Classification of Restaurant Images

In this paper, we present a novel algorithm for multi instance multi label classification of attributes in user submitted restaurant photos. The features are extracted from images using a seven layer convolution neural network architecture which are then represented using the bag of words model with earth mover's distance as a metric to compute the distance between each bag. For classification, we employ a custom kernel support vector machines that is trained for each attribute to obtain binary classifiers which are then used to predict the labels for an unknown business. However, central to our approach is the mining of association rules between different labels to further increase the accuracy of the proposed system. The method is evaluated on the Yelp Kaggle database and the performance is measured on the basis of F1 score. The experimental results prove that our algorithm is effective in classification of restaurant images.