音乐声的音色感知特征与图像的纹理特征的关联研究

Research on Association between Timbre Perception Features of Musical Sound and Texture Features of Image

本文主要研究音乐声的音色感知特征与图像的纹理之间的关联关系并建立音色与纹理的匹配模型.首先对Brodatz纹理库中的图片提取纹理特征,然后利用纹理特征对纹理图片进行系统聚类和多维尺度分析,挑选出18张用来做主观评价实验的纹理图片.通过主观评价实验,得到音色感知特征值以及音色和纹理的匹配关系数据,并基于此数据分别计算了匹配(或不匹配)的音色感知特征纹理特征之间的皮尔逊相关系数,相关系数绝对值最高达0.75,证明了二者之间存在强烈关联.最后使用逻辑回归、随机森林、多层感知神经网络对音色感知特征和纹理之间的匹配关系进行预测,得到了准确率较高的预测模型,说明通过音色感知特征和纹理特征预测二者匹配关系是可行的.

This research studies the association between the timbre perception feature of music and the texture feature of images.Moreover,the matching model of timbre and texture is established.First of all,texture features are extracted from images in Brodatz texture library.Then the system clustering and multidimensional scaling are carried out for selecting texture images from Brodatz texture library,and 18 texture images are selected for subjective evaluation experiment.Through the subjective evaluation experiment,the perception feature value of timbre and the matching relationship of timbre and texture are obtained.Based on this data,the Pearson correlation coefficient about matching(or mismatching)timbre-texture feature pairs is calculated,the absolute value of the correlation coefficient peaked at 0.75,which verifies the strong association between them.Then,logistic regression,random forest and multilayer perceptual neural network are used to predict the matching relationship between texture and timbre,and a prediction model with high precision is obtained,which proves the feasibility of predicting the matching relationship between timbre perception features and texture features.