Article: Multimodal deep neural network inference for design colour emotion quantification Journal: International Journal of Information and Communication Technology (IJICT) 2026 Vol.27 No.96 pp.63 - 83 Abstract: Existing design tools can recommend harmonious colour palettes yet fail to quantify the emotions a scheme evokes, leaving designers to rely on intuition rather than measurable feedback. This paper introduces a multimodal deep neural framework that continuously measures the affective qualities of design colours. The model fuses hierarchical visual features from design images with semantic intentions extracted from designer descriptions via a cross-attention mechanism. A perceptual colour-difference regularisation loss ensures physically plausible predictions, enforcing smooth emotional transitions consistent with human perception. Tested on three datasets, our method achieves a root-mean-square error of 0.0873 for emotion regression, a relative gain of 16.2% over the strongest baseline, and yields ratings statistically indistinguishable from those of professional designers. The framework bridges colour physics and design cognition, enabling emotion-aware design assistance. Inderscience Publishers - linking academia, business and industry through research

Title: Multimodal deep neural network inference for design colour emotion quantification

Authors: Wenjie Wang

Addresses: College of Art and Design, Lanzhou City University, Lanzhou, 730070, China

Abstract: Existing design tools can recommend harmonious colour palettes yet fail to quantify the emotions a scheme evokes, leaving designers to rely on intuition rather than measurable feedback. This paper introduces a multimodal deep neural framework that continuously measures the affective qualities of design colours. The model fuses hierarchical visual features from design images with semantic intentions extracted from designer descriptions via a cross-attention mechanism. A perceptual colour-difference regularisation loss ensures physically plausible predictions, enforcing smooth emotional transitions consistent with human perception. Tested on three datasets, our method achieves a root-mean-square error of 0.0873 for emotion regression, a relative gain of 16.2% over the strongest baseline, and yields ratings statistically indistinguishable from those of professional designers. The framework bridges colour physics and design cognition, enabling emotion-aware design assistance.

Keywords: colour emotion quantification; multimodal deep learning; design cognition; physical constraints; affective computing.

DOI: 10.1504/IJICT.2026.156194

International Journal of Information and Communication Technology, 2026 Vol.27 No.96, pp.63 - 83

Received: 13 May 2026
Accepted: 18 Jun 2026
Published online: 07 Sep 2026 *

Title: Multimodal deep neural network inference for design colour emotion quantification

Keep up-to-date