图像视点预测模型及设计图的注意力影响因素研究  

RESEARCH ON IMAGE VIEWPOINT PREDICTION MODEL AND ATTENTION INFLUENCING FACTORS OF DESIGN DRAWING

作　　者：侯士江[1] 侯英[1] 冯希[1] 王兆琦 HOU SHIJIANG;HOU YING;FENG XI

机构地区：[1]燕山大学艺术与设计学院

出　　处：《设计》2022年第14期134-136,共3页Design

基　　金：河北省社会科学基金年度项目(HB18YS042)。

摘　　要：为了满足设计图的细粒度视点预测要求,开发了基于深度学习的视点预测模型,以实现热点图生成、特征要素辨识及设计方案的交互式即时检测,并讨论了注意力影响因素。引入显著图来模拟视觉注意力分配机制,提出基于全卷积神经网络的图像视点预测模型(IVPM),克服了眼动仪测试的诸多限制,模型在图形设计重要性(GDI)数据集上训练后具有出色的时间性能,实验验证了图像的低层级属性是设计注意力的主要影响因素。IVPM可以应用于自然图像、海报设计、包装设计、产品设计以及界面设计等领域,对相关设计工作具有一定的参考价值。In order to meet the fine-grained viewpoint prediction requirements of design drawings,a viewpoint prediction model based on deep learning is developed to realize heat map generation,feature element identification and interactive real-time detection of design schemes,and the influencing factors a re discussed.The saliency map is introduced to simulate the visual attention allocation mechanism,and an image viewpoint prediction model(IVPM) based on full convolution neural network is proposed,which overcomes many limitations of eye tracker testing.The model has excellent time performance after training on graphic design importance(GDI) data set.Experiments show that the low-level attribute of the image is the main influencing factor of design attention.IVPM can be applied to natural image,poster design,packaging design,product design,interface design and other fields,and has a ce rtain reference value for relevant design work.

关 键 词：注意力管理 眼动跟踪 视点预测 产品设计 深度学习 

分 类 号：TB472[一般工业技术—工业设计]