Automatic image annotation using a hybrid engine

Humans can view an image and immediately determine what the image is trying to convey. While this may be an easy event for humans, it is still considerably difficult for a computer to understand of its own accord. The challenge broadly lies in developing an automatic process to complement and supplant human visual and neural systems. In this paper, we address the core issue of imparting an image the ability to caption itself automatically. We propose a hybrid engine that utilizes a combination of feature detection algorithms coupled with context-free grammar to create a model that serves to semantically and logically describe an image in its entirety. Our hybrid engine model has an F1 score of 94.33% and a unigram score of 75% when evaluated on a novel dataset trained on human-annotated images.