Date of Award


Document Type

Thesis (Undergraduate)

Department or Program

Department of Computer Science

First Advisor

Emily Cooper


People can recognize the context of a scene with just a brief glance. Visual information such as color, objects and their properties, and texture are all important in correctly determining the type of scene (e.g. indoors versus outdoors). Although these properties are all useful, it is unclear which features of an image play a more important role in the task of scene recognition. To this aim, we compare and contrast a state-of-the-art neural network and GIST model with human performance on the task of classifying images as indoors or outdoors. We analyze the impact of image manipulations, such as blurring and scrambling, on computational models of scene recognition and human perception. We then create and analyze a measure of local-global information to represent how each perceptual system relies on local and global image features. Finally, we train a variety of neural networks on degraded images to attempt to build a neural network that emulates human performance on both classificaton accuracies and this local-global measure.


Originally posted in the Dartmouth College Computer Science Technical Report Series, number TR2017-820.