Department or Program


Second Department or Program

<--Please Select Department-->


When we look at the world around us we are able to effortlessly categorize scenes, but it is still unclear what mechanisms we use to do so. Categorization could be driven by objects, low-level features, or a mixture of both. This study investigated the ways in which diagnostic objects (those found nearly exclusively in one scene category) contribute to scene categorization. It paired Electroencephalography (EEG) with machine learning classification to provide detailed temporal information about when categorization occurs. While recording EEG, participants categorized real-world photographs as one of three indoor scene types (bathroom, kitchen, office). They were shown either original images or versions where diagnostic or random objects had been obscured via localized Fourier phase randomization. EEG voltages and the independent components (ICs) of a whole brain independent component analysis (ICA) were used as feature vectors for a linear support vector machine (SVM) classifier to determine time-resolved accuracy. There were no significant differences in decoding accuracy between categories or between diagnostic and random conditions. Poor classifier performance is likely due to a lack of power, or overfitting of the model. It could also reflect unclear EEG-based neural correlates of each scene type due to the inherent similarities in the categories. While the lack of significant decoding makes it difficult to make strong conclusions about the role of diagnostic objects in visual scene categorization, this study addresses important considerations for pairing EEG with decoding techniques and highlights some of the broader difficulties of isolating distinct features of visual scenes.

Level of Access

Open Access

First Advisor

Greene, Michelle

Date of Graduation


Degree Name

Bachelor of Arts

Number of Pages


Components of Thesis

One PDF document.

Open Access

Available to all.