摘要
为快速、批量化检测服饰品的视觉兴趣度,依据人眼视觉注意模型,利用自底向上的视觉显著性检测方法描述服饰图像的视觉显著性。提取服饰图像的亮度、色彩、纹理方向3个底层特征,利用多尺度的分解方法构建出图像的多特征通道;采用归一化合并的方式对特征显著图进行滤波;对形成的3类特征图进行融合,根据显著程度绘制视觉热点图及亮度图,从而划分图像的视觉显著区域,并进一步提出了视觉覆盖率和分散度2个指标,用以对服饰图像的视觉显著性差异进行量化和评价。对常见色块、不同纹理织物、服装及配饰图视觉显著度进行实验检测,结果表明,本文方法可有效地描绘出图像视觉显著区域,快速、客观地评定出服饰图像的视觉显著度。
In order to fastly and batch detect the visual interestingness of clothing and accessory images,the bottom-to-top visual saliency detection method was utilized to describe the visual saliency based on human visual attention model. Firstly,the bottom characteristics including brightness,color and texture of fashion image were extracted to construct multiple feature channels. Besides,combining normalized method was used to filter the feature image. Finally,the constructed three feature maps were fused and the visual focus area and significant bright map layer were converged. In the experiment,the visual saliency of color blocks map, different texture fabric images, clothing and accessory images were detected. The results show that the proposed algorithm effectively and objectively extracts the salient region and presents the significant degree.
出处
《纺织学报》
EI
CAS
CSCD
北大核心
2018年第3期126-131,共6页
Journal of Textile Research
基金
国家自然科学基金项目面上项目(61702460
71373041)
中央高校基本科研业务费专项资金项目(CSQ13038)
浙江省教育厅科研项目(Y201738456)
湖北省民宗委文化研究项目(QSZ13009)
浙江理工大学科研启动基金项目(17072067-Y)
关键词
服饰设计
视觉显著性
热点图
客观评介
fashion and accessory design
visual saliency
hot map
objective evaluation
