基于梯度归一化的模糊梯度特征提取方法

Fuzzy gradient feature extraction method based on gradient normalization

提出一种应用于手写字符识别的基于梯度归一化模糊梯度特征提取方法.首先计算原图像的梯度;然后基于一定的归一化函数,得到归一化梯度;最后,基于归一化梯度,构建模糊梯度特征向量.针对归一化函数,提出了分段线性归一化函数,它能够有效减小类内样本分散度,同时具有计算简单高效的优点.针对梯度特征向量构建方法,提出了模糊梯度特征,改进了普通梯度特征向量的构建方法,提高了梯度特征吸收手写字符形变的能力.

A fuzzy gradient feature extraction method based on gradient normalization applied in handwritten character recognition image is calculated first; then, normalized gradient is obtained by use of certain normalization function ; last, on the basis of normalized gradient, fuzzy gradient feature vector is constructed. With regard to normalization function, a piecewise linear normali- zation function is proposed. It can effectively alleviate the divergence of samples within-class and as well as has merits of high computation efficiency. With regard to gradient feature vector construction method, fuzzy gradient feature is proposed. It improves common gradient feature vector construction method and its capability of absorbing variations of handwritten character is enhanced.