Investigating the Gestalt Principle of Closure in Deep Convolutional Neural Networks

TL;DR

The findings reveal a performance degradation as the edge removal percentage increases, indicating that current models heavily rely on complete edge information for accurate classification.

Abstract

Deep neural networks perform well in object recognition, but do they perceive objects like humans? This study investigates the Gestalt principle of closure in convolutional neural networks. We propose a protocol to identify closure and conduct experiments using simple visual stimuli with progressively removed edge sections. We evaluate well-known networks on their ability to classify incomplete polygons. Our findings reveal a performance degradation as the edge removal percentage increases, indicating that current models heavily rely on complete edge information for accurate classification. The data used in our study is available on Github.

Figures (2)

• Figure 1: Image examples used in the (a) training set (b) test set. Each column indicates a different type of polygon (i.e., triangle, square, pentagon, hexagon, and so on). In (a), the top two rows show polygons with different $\theta_{global}$, the middle two rows feature different background colours, and the bottom two rows illustrate various positions. In (b), each row shows polygons with varying percentages of removal from each side. The settings for $\theta_{global}$, background colours, and centre positions of the polygons in the test set are the same as those used in the training set, although these variables are not depicted in (b).
• Figure 2: The percentage of correctly classified examples in the test set versus (a) the removal percentage for each model in each side of a polygon and (b) the number of vertices at 10% removal percentage.