Estimation of Effective Commercial Signboard Design Scores Based on Visual Perception

Nowadays, numerous commercial signboards, billboards and banners can be found throughout urban areas. Among these, store signboards are one of the most important structuring elements for constructing pleasant urban landscapes. Since effective store signboards are usually informational and visually appealing, it is essential to consider both the artistic and informational aspects when designing store signboards. In this paper, we propose a new computational model that evaluates the effectiveness of commercial signboard design in terms of aesthetics, memorability, and saliency as criteria. Our proposed model consists of two steps: crowdsourcing and prediction. In the crowdsourcing phase, we conduct multi-modal cascaded test to evaluate the effectiveness of signboards. In the prediction phase, we propose a modified Capsule Network (CapsNet) for computing the effectiveness score of a commercial signboard. Experimental results demonstrate that our proposed system shows reasonable performance with a RMSE of 0.0721 in terms of predicting effectiveness score for a given commercial signboard.