基于机器学习的墙纸纹理风格学习与场景推荐

Wallpaper Texture Style Learning and Scene Recommendation Based on Machine Learning

摘要: 室内装饰性的墙纸纹理具有色彩多样、图案丰富的特点，包含现代风格、巴洛克风格、奢华风格等的多种不同风格，不同风格的墙纸适用于卧室、客厅、浴室等多种不同的现实场景。墙纸纹理的风格变化趋势引起了许多墙纸设计公司、墙纸售卖者的广泛关注。设计师需要识别不同风格墙纸纹理，并针对不同的房间场景，为客户推荐合适的纹理。此过程耗费较多人力和时间，我们希望可以使用计算机实现自动墙纸纹理风格标注与推荐。本文构建了包含多种墙纸纹理风格的图像数据集，在计算机视觉和机器学习方法的基础上，使用特征学习与SVM分类器构建了墙纸纹理的风格学习和应用场景推荐系统，实现了针对输入墙纸纹理图像进行风格分析预测，并对该墙纸纹理所适用的场景进行了推荐。

Abstract: Indoor decorative wallpaper textures have the characters of full color printing and abundant pattern, including modern style, baroque style, luxury style and some different styles; the wallpapers with different styles are suitable for different scenes such as bedroom, living room and bathroom. The variation of wallpaper texture styles are drawing attention from wallpaper design companies and wallpaper sellers. The designers are supposed to recognize the style of different wallpaper textures, and recommend suitable textures to costumes according to the scenes, which required more people and time; we hope to realize wallpaper texture style labeling and recommendation automatically. In this paper, a database of wallpaper textures with different styles is collected. Based on computer vision and machine learning methods, a wallpaper texture style learning and scene recommendation system is established with feature learning and support vector machine classifier, which realized the process of style prediction and scene recommendation with input wallpaper textures.