摘要： 通过单幅图像数据的高维化,将单幅图像转化为高维空间中的数据集合,对其进行非线性降维,寻求其高维数据流形本征结构的低维表示向量,将其作为图像数据的特征表达向量。从而将高维图像识别问题转化为特征表达向量的识别问题。大大降低了计算的复杂程度,减少了冗余信息所造成的识别误差。通过指纹图像的实例说明,将非线性降维方法应用于图像数据识别问题,在实际中是可行的,在计算上是简单的,可大大改善常用识别方法的效能。

关键词: 图像数据, 非线性降维, 本征结构

Abstract: By transforming each image to high-dimension data set and the nonlinear dimension reduction, the low-dimension expression on the intrinsic structure of the data manifold is acquired, which can be used to represent the image sufficiently. Consequently, the recognition will be translated into the low-dimension expression’s recognition. That will greatly reduce the calculative complexity and the error, which comes from the data redundancies. Finally, the example of the fingerprints shows that it is feasible and valid to image recognition. It can greatly improve the nearest neighbor’s recognition results. Furthermore, the insensitivity to image is convenient to recognize.

中图分类号: 

• TP391.41