Visual Categorization: Basic v Super/Sub-ordinate levels

SE367 Project - Introduction to Cognitive sciences

Bhuwan Dhingra

Mentor: Prof. Amitabha Mukherjee

Objective: To study the hierarchy of object categorization using a computational model for vision.

Natural Categories: Rosch and her colleagues showed in the 1970s that there are three levels of categorizations - basic, super-ordinate and subordinate (see here for details).

Abstract: A computational model to study visual categorization of objects is presented. It is widely accepted in the domain of cognitive sciences that there exist three levels of categorization - super-ordinate, basic and the subordinate. Rosch and her colleagues showed in the 1970s that out of these the basic level is accessed first. Recent studies, however, have pointed to the evidence that in a rapid visual processing task super-ordinate categories dominate. In this work I attempt to study this hierarchy of categorizations using an object recognition classifier. A bag-of-features model is used for feature extraction from object images, followed by k-means clustering to implement categorization. Results show that super-ordinate categories are indeed favoured over basic and subordinate ones. The role of expertise in determining the categories is also studied.

Dataset: Following hierarchy of object categorizations used:

For each sub-ordinate category 30 images were downloaded from Google image search. Hence, the taxonomy had a total of 240 images. Some examples include:

           

           

Results: The different levels of categorization were evaluated on four different performance metrics - Rand Index (RI), Purity, Normalized Mutual Information (NMI), Silhouette Index. There variation with the Peak Threshold, a parameter controlling the number of detected key-points per image was studied. As the peak threshold increases the number of key-points decreases.

Important Links -

1. Paper review.
2. Proposal.
3. Final presentation.
4. Final report.
5. MATLAB Code.

Main References -

[1] Rosch, E. (1973). Natural categories. Cognitive Psychology.
[2] Rosch, E., Mervis, C., Gray, W., Johnson, D., & Boyes-Braem, P. (1976). Basic objects in natural categories. Cognitive Psychology.
[3] Marc, J.M.M., Joubert, O.R., Nespoulous, J.L. & Fabre-Thorpe, M (2009). The time-course of visual categorizations: you spot the animal faster than the bird. PLoS one.
[4] Johnson, K.E., Mervis, C.B. (1997). Effects of varying levels of expertise on the basic level of categorization. Journal of Expert Psychology.
[5] Jiang, Y.G., Ngo, C.W., Yang, J. (2007). Towards Optimal Bag-of-Features for Object Categorization and Semantic Video Retrieval. CIVR'07.