Abstract
In recent years, deep learning has transitioned from solving traditional algorithmic problems to enhancing product emotional design. This paper explores the integration of deep learning and affective technology in Augmented Reality (AR) interactive picture books. By transforming the reading process from static to dynamic, the proposed framework aims to enrich visual stimulation and increase interactivity. Based on a three-level emotional design model, image emotion labeling results were used to train a deep neural network. The resulting model demonstrates a significant improvement in emotional recognition accuracy compared to traditional algorithms, addressing the "semantic gap" in current AR marketing and meeting children's emotional reading needs.
Introduction
As digital media increasingly competes for children's attention, traditional paper-based picture books face challenges in engagement. While Augmented Reality (AR) offers a potential solution, many current implementations are dismissed as "marketing gimmicks"—3D models that lack a meaningful connection to the narrative's emotional core. The primary obstacle is the semantic gap: the difficulty for computational systems to interpret high-level emotional concepts like "trust" or "comfort" from low-level visual data like pixels and shapes. This research proposes bridging this gap by combining deep learning architectures with established emotional design theories.
Theoretical Framework: Emotional Design
The foundation of this study rests on Donald Norman’s three-level theoretical model of emotional design. This model categorizes user experience into three distinct layers:
- Instinctive Design: Focuses on immediate sensory impact, such as color, texture, and form.
- Behavioral Design: Relates to the functionality, usability, and the joy of interaction.
- Reflective Design: Concerns deep consciousness, cultural meaning, and long-term psychological impact.
By applying this framework, designers can move beyond superficial visual effects to create AR content that resonates with a child's psychological development.
Methodology and Emotional Classification
To automate the emotional analysis of picture book illustrations, the researchers selected the VGGNet deep learning architecture. VGGNet was chosen for its superior generalization capabilities in computer vision. Its use of small (3x3) convolutional kernels allows the network to capture nuanced "sensory fields" essential for identifying subtle emotional cues.
Dataset and Labeling
A custom "Preliminary Design Gallery" was constructed, consisting of 2,048 images sourced from award-winning product designs. These images were selected for their clear emotional characteristics and labeled by design professionals across 11 categories: Easy, Comfort, Simple, Trust, Softness, Happy, Noble, Surprise, Ordered, Lovely, and Belonging.
The dataset was split into 1,500 training images and 548 testing images. The labeling process revealed that "Easy" (22.7%) and "Comfort" (12.15%) were the most prevalent emotions, suggesting a current industry focus on the "Behavioral" level of design.
Results and Discussion
The VGGNet model achieved a single-label prediction accuracy of 61.02%. This represents a notable advancement over traditional machine learning algorithms, which typically range between 50% and 60% for similar emotional classification tasks.
Key findings include:
- Deep Learning Efficacy: While human observers often find it difficult to articulate why an image feels "noble" or "soft," the deep neural network successfully "learned" these patterns through backpropagation and weight optimization.
- The Reflective Gap: The "Reflective" layer (e.g., "Belonging") was the least represented in the dataset (1.71%). This highlights a significant opportunity for future interactive media to deepen its emotional and cultural resonance.
Conclusion
The integration of AR virtual implantation with deep learning-based emotional recognition offers a path toward more meaningful interactive picture books. By moving beyond simple 3D overlays and aligning digital content with the emotional arc of a story, creators can better stimulate children's creativity and engagement. Future research will focus on expanding the image dataset and specifically targeting the "Reflective" layer of emotional design to foster long-term psychological benefits.
Summary updated