Research on Cellphone Appearance Image Models Based on BP

Yan Zhou, Yizhou Yun, Ping YangAdvances 2015 · DOI

Abstract

Kansei engineering provides a new idea for perceptual quantitative analysis. This paper targets in quantitative description the relationship between cellphone appearance design and consumer subjective felling, eventually builds a mathematical model through BP neural network. This research start with collecting a large number of perceptual image vocabularies, then transforms subjective assessment to quantitative evaluation by SD method and builds BP neural network mode. Thus analyzes the result of BP Model, the foundation is established by summary of the relationship between product appearance and objective feeling. It can be used to know the rule of how modeling elements affect perceptual image through the complete BP mode. This research provides a scientific method for engineer to process emotional design problem.

Introduction

In the field of industrial design, the "look and feel" of a product is often treated as a subjective "black box." While designers may aim for a specific emotional response—such as "professionalism" or "delicacy"—there is rarely a mathematical framework to guarantee that consumers will perceive the product as intended. This research addresses the core challenge of Kansei Engineering (KE): converting subjective, perceptual factors into objective, visible data. By leveraging Back-Propagation (BP) neural networks, the authors aim to move away from trial-and-error design toward a predictive model where specific geometric changes result in predictable shifts in consumer perception.

Methodology and Data Collection

The researchers employed a multi-step pipeline to bridge the gap between geometry and emotion, utilizing a dataset of 20 representative mobile phone modeling samples.

1. Feature Deconstruction

Each cellphone sample was deconstructed into six primary modeling features, categorized into continuous and discrete variables:

  • Continuous: Edge horn shape, screen size, and thickness.
  • Discrete: Side section style, cross-section style, and button shape.
Example of a mobile phone sample used for feature deconstruction, showing front, back, and side profiles.

2. Semantic Differential (SD) Method

To quantify subjective feelings, the researchers curated an initial pool of 264 perceptual words, narrowing them down to six representative bipolar adjective pairs: durable-delicate, rigid-soft, beautiful-ugly, professional-amateur, unique-ordinary, and simple-tedious. Participants rated the 20 phone samples on these scales to create a quantitative ground-truth dataset.

A secondary sample illustrating a different design language (e.g., curved back and distinct button placement) used to train the model's sensitivity.

The BP Neural Network Model

A three-layer artificial neural network was constructed to map physical features to emotional responses:

  • Input Layer: 15 nodes representing the deconstructed physical features.
  • Hidden Layer: 10 nodes, utilizing a Tansig (hyperbolic tangent sigmoid) activation function.
  • Output Layer: 6 nodes representing the adjective pairs, utilizing a Purelin (linear) function to map correctly to the SD scale of -2 to 2.

The network was trained using MATLAB’s neural network toolbox and the Levenberg-Marquardt (LM) algorithm for 1000 epochs to minimize the error between predicted and actual human ratings.

Key Results and Insights

The model demonstrated high accuracy, converging with an absolute error of less than 0.1. This suggests that "feeling" is not entirely random and that non-linear relationships exist between a phone's geometry and its perceived image.

Through "influence rule analysis," the researchers identified specific design correlations:

  • Side section style 'a2' had the highest positive impact on making a phone appear "beautiful" and "professional."
  • Side section style 'a4' was perceived most negatively, correlating with "ugly" or "amateur" ratings.
  • Side section style 'a3' was the primary driver for the "rigid-soft" perception.

Conclusion

This study provides a scientific method for engineers to process emotional design problems. By holding variables constant and adjusting single features, designers can see how changes ripple through emotional categories. However, the study is limited by a small sample size (20 phones) and a lack of demographic diversity in the focus groups. Future work should incorporate more granular geometric elements, such as point and line structures, to handle increasingly complex product designs.

Summary updated

Abstract

Kansei engineering provides a new idea for perceptual quantitative analysis. This paper targets in quantitative description the relationship between cellphone appearance design and consumer subjective felling, eventually builds a mathematical model through BP neural network. This research start with collecting a large number of perceptual image vocabularies, then transforms subjective assessment to quantitative evaluation by SD method and builds BP neural network mode. Thus analyzes the result of BP Model, the foundation is established by summary of the relationship between product appearance and objective feeling. It can be used to know the rule of how modeling elements affect perceptual image through the complete BP mode. This research provides a scientific method for engineer to process emotional design problem.
BibTeX
@inproceedings{zhou2015research,
  author = {Yan Zhou and Yizhou Yun and Ping Yang},
  title = {{Research on Cellphone Appearance Image Models Based on BP}},
  booktitle = {Advances},
  year = {2015},
  doi = {10.2991/amcce-15.2015.106},
}
Figure