一种新的彩色图像降维方法

A New Nonlinear Dimensionality Reduction for Color Image

基于内容的图像检索(CBIR)是图像检索的重要分支,而基于颜色的特征提取是CBIR的常用方法之一.如果对图像颜色的特征数提取过多、维数过大,则不利于对图像的快速匹配.本文将图像的色彩直方图作为输入向量,然后采用局部线性映射(LLE)算法对原始数据进行降维,并分别在4种色彩空间下对降维后的彩色图像进行分类.实验证明,在处理非线性数据降维时,LLE较主成分分析(PCA)具有明显的优势.

Content Based Image Retrieval (CBIR) is an important component in the context of image, and the color based feature extraction is often used in CBIR. Large number of image features is an obstacle for fast image matching. The chromaticity histogram was extracted as the input features and then the locally (linear) embedding (LLE) algorithm was used to reduce the dimensionality. At last, the color images in four (color) spaces was classified by simple classifiers. The results prove that the LLE outperforms the principal component analysis (PCA) while dealing with nonlinear data in the task of dimensionality reduction.