基于图像核函数的图像目标识别技术研究

核方法已经广泛应用于模式识别的各个领域,但是传统的核函数仅能接受一维矢量作为输入数据,对二维图像数据需要进行一定的预处理.本文从核函数构造的角度出发,以RBF核函数为结构原型,定义了一种输入量为图像数据的新型核函数.从形式上看,本核函数仅在输人数据的类型上有所不同,其他方面完全一致,经典的基于核方法的分类器算法都可以使用本核函数.对UCI数据实验结果表明,本核函数在多种图像识别应用中均有良好的效果.

Image Targets Recognition based on Image Kernel

Kernel methods have been widely used in every area of pattern recognitions.But traditional kemel functions can onlv accept 1-D vectors as their input,2-D image data must be preproeessed.This paper works on the constructing of kernel functions.and proposes a new kernel function which can accept 2-D image data as input directly.based on the traditional RBF kernels.This new kernel function is the same as traditional kernels except on inputs, and can be used in the common methods.such as SVM and SVDD.Theexperiments on UCI datasets show that,the classifier based on our kernel function can have good performance.