Unsupervised computer vision

Here’s the abstract of a working project. We leverage on computer vision to measure similarity across a corpus of product design images. We further validate the measure, and use it to show that aesthetic knowledge is geographically localized.


Computer vision (CV) enables computers to “see” and “understand” images. This paper provides guidelines for performing CV in management research and then elaborates these guideline with a detailed application of the Structural Similarity Index Measure (SSIM) to quantify the visual similarity across product designs. We first apply SSIM to measure the visual similarity of designs over a corpus of 611,578 US design patent images granted from 1976 to 2020, and validate the measure with a validation survey involving 382 online observers. Finally, using the validated similarity measure, we provide new and rigorous evidence that aesthetic knowledge spillovers are geographically localized. We discuss how the ability to leverage CV to measure similarity over images opens new avenues of research, and we provide open-access code and data.

Tian Chan
Tian Chan
Assistant Professor (Information Systems and Operations Management)

Scholar researching and teaching in the areas of operations and innovation management.