一种基于相关性分析与对数搜索聚类的跨媒体检索方法

A cross media retrieval method based on correlation analysis and logarithmic search clustering

为了在不需人工标记的多媒体数据集上进行跨媒体检索,避免不同媒体之间的语义提取与关联时具有的局限性等问题,首先提出了一种基于相关性分析的音频与图像间的关系数据集的收集机制,然后提出了一种基于对数搜索思想的聚类方法用于跨媒体检索模型的训练,最后提出了一种基于关联性分析的跨媒体检索方法。实验结果表明,所提方案的音频检索图像的第一结果的平均相关度为82.92%,而图像检索音频的第一结果的平均相关度为61.25%,能够较好地解决当前环境下的图像、音频之间的跨媒体搜索需求。

To carry out cross-media retrieval on multimedia datasets without manual labeling,and to avoid the limitations of semantic extraction and association between different media,a correlation between audio and image based on correlation analysis is proposed.Based on the collection mechanism of relational data sets,a clustering method based on logarithmic search idea is proposed for the training of cross-media retrieval models.Then,a cross-media retrieval method based on relevance analysis is proposed.The experimental results show that the average correlation of the first result of the audio search image of the proposed scheme is 82.92%,and the average correlation of the first result of the image retrieval audio is 61.25%,which can better solve the cross-media search needs between image and audio in the current environment.