Contents

-

Study on the Relevance of Jiangnan Garden Design Elements and Aesthetic Context Based on Visual Recognition

Wenjing Liang1, Yijing Chen2, Nadia Binti Mohd Nasir3
1 School of Art, Shanghai Zhongqiao Vocational and Technical University, Shanghai, 201514, China
2 School of Humanities & Art, Bengbu College of Technology and Business, Bengbu, Anhui, 233000, China
3Faculty of Creative Industry and Communication, City University Malaysia, Kuala Lumpur, 50000, Malaysia

Abstract

Jiangnan gardens have become a valuable cultural heritage of China with its elegant garden style. The article proposes a binocular visual recognition system by analyzing the composition of the garden spatial elements and performing feature fusion based on scene-driven coefficients. Ablation experiments are conducted on each part of the constructed data enhancement framework for generating the design of the Jiangnan garden plan, which is applied to generate a set of high-quality datasets and apply the data to image segmentation for generating the design of the Jiangnan garden. The algorithm training is carried out by applying the generated design plan dataset. On this basis, the data from the actual Jiangnan garden research and the spatially quantized feature data are used to do the correlation analysis between the design elements and the aesthetic mood. The data enhancement framework constructed in this paper improves the IOU of ST elements to 0.537, and the average intersection and merger ratio MIOU is 0.389. It shows that the data evaluation framework based on visual recognition is suitable for the study of plan generation of Jiangnan gardens. The correlation coefficients of connection value, spatial control value, average depth value, and integration degree regarding aesthetic context with the data of Jiangnan garden design elements are 0.173, 0.301, -0.278, and 0.325, respectively, which indicate that there is a significant correlation between all of them.

Keywords: jiangnan garden, visual recognition, binocular vision system, scene-driven, multiplexed feature fusion