Visual contextual perception and user emotional feedback in visual communication design.

Background: With the advent of the information era, the significance of visual communication design has escalated within the realm of increasingly prevalent network applications. Addressing the deficiency observed in prevailing sentiment analysis approaches in visual communication design, which predominantly leverage the holistic image information while overlooking the nuances inherent in the localized regions that accentuate emotion, coupled with the inadequacy in semantically mining diverse channel features.

Methods: This paper introduces a dual-attention multilayer feature fusion-based methodology denoted as DA-MLCNN. Initially, a multilayer convolutional neural network (CNN) feature extraction architecture is devised to effectuate the amalgamation of both overall and localized features, thereby extracting both high-level and low-level features inherent in the image. Furthermore, the integration of a spatial attention mechanism fortifies the low-level features, while a channel attention mechanism bolsters the high-level features. Ultimately, the features augmented by the attention mechanisms are harmonized to yield semantically enriched discerning visual features for training sentiment classifiers.

Results: This culminates in attaining classification accuracies of 79.8% and 55.8% on the Twitter 2017 and Emotion ROI datasets, respectively. Furthermore, the method attains classification accuracies of 89%, 94%, and 91% for the three categories of sadness, surprise, and joy on the Emotion ROI dataset.

Conclusions: The efficacy demonstrated on dichotomous and multicategorical emotion image datasets underscores the capacity of the proposed approach to acquire more discriminative visual features, thereby enhancing the landscape of visual sentiment analysis. The elevated performance of the visual sentiment analysis method serves to catalyze innovative advancements in visual communication design, offering designers expanded prospects and possibilities.