基于内容的图像情感语义检索研究

摘要

基于内容的图像检索是使用图像的底层视觉特征对图像进行检索，使检索结果在视觉角度上尽可能相似。但能否通过图像的底层特征来准确体现人对图像的视觉感知（即图像的情感语义）有待于进一步的探索。首先构建检索性能较好的基于内容的图像检索系统，并针对分类标准不同的两类图像库进行多次实验。实验证明，图像的情感语义无法通过单一的图像底层特征描述。%Content-based image retrieval means retrieving images by image low level visual features so that the retrieved results are as simi-lar as possible from the visual perspective.However it requires further exploration whether image low lever features can accurately represent visual perception,or image affective semantics,of human upon image.The paper first of all constructs content-based image retrieval system with better retrieval performance;next it carries out a number of experiments upon two image databases that are classified by different criteria. Experiments prove that image affective semantics can't be described by single image low layer features.