Abstract: This paper presents an image semantic description in high-level information. In order to explain and calculate the image semantics, the feature vectors and the semantic partition are constructed in structure related. And also the selection of vector space is discussed. In the vector space, an idea is given for the selection of minimum vector dimensions which is being proved by discriminant. From the vectors selection, the structure of feature projection and the formulation can effective to calculate the data from low-level to high-level in image retrieval and recognition. The principles and the experiments show the thoughts can help others in semantic retrieving.

Key words: image semantics, extraction of feature, semantic projection

摘要： 从高级信息的角度来描述图像语义，建立图像语义的特征矢量空间和语义划分的结构关系，实现图像与语义值的结构表达。为了有效地获取语义特征值表达，给出了图像语义特征空间选择与最小判别方法，构建了底层特征到高层语义的映射结构与计算表达式，并将特征值应用于图像检索。原理方法和实验数据表明该方法对图像检索具有积极意义。

关键词: 图像语义, 特征抽取, 语义映射

CLC Number: 

• TP319.4