Original scientific paper
https://doi.org/10.20532/cit.2023.1005695
A Visual Cortex-Attentive Deep Convolutional Neural Network for Digital Image Design
Lei Zheng
; Shanghai Lida University, Shanghai, China
Abstract
With the proliferation of advanced visualization techniques in visual communication, enhancing digital image quality remains a persistent challenge. This study presents a sophisticated Convolutional Neural Network (CNN) model to optimize image processing. The model incorporates a multi-stage architecture attentive to biological visual pathways. Inter-subnetwork connections enable integrated feature learning, guided by adaptive weighting of luminance, color, orientation, and edge maps. Spatial and channel attention modules further enrich feature interplay. When evaluated on the LIVE 3D Phase dataset, the approach demonstrates marked improvements, with saliency maps closely mirroring human visual perception. Pearson Correlation Coefficient and Histogram Intersection metrics exceed conventional models, at 0.6486 and 0.7074 respectively. Testing across distortion types reveals strong agreement with subjective rankings, confirming the model's effectiveness. By combining automated feature extraction with insights from visual cortex mechanisms, this bio-inspired CNN framework significantly enhances image optimization and quality. The scalable approach provides a foundation for next-generation computer vision and machine learning applications.
Keywords
HSV, CNN, visual communication, bilinear combination
Hrčak ID:
313205
URI
Publication date:
8.1.2024.
Visits: 116 *