针对具有较高识别率,但计算量大,对噪声比较敏感的Zernike矩特征,提出一种结合不可分小波四通道分解方法与Zernike矩的特征提取方法.在对预处理后的文字图像进行一层不可分小波分解后,只取其中的低频子图,然后再计算该低频子图的Zernike矩,将其作为原图像特征,最后依据该特征进行识别分类.实验结果表明,对比于单纯计算原图像Zernike矩作为特征的方法,所提方法拥有相对较小的计算量,较好的抗噪性以及更高的识别率.同时,对比于最新发展起来的卷积神经网络识别法,所提方法拥有与其几乎相同的识别率以及相对较小的计算量.%Aiming at the characteristics of Zernike moments with high recognition rate,high computational complexity and sensitive to noise,a four-channel decomposition method based on indivisible wavelet and Zernike moments feature extraction method is proposed.After the preprocessed text images were decomposed by the indivisible wavelet,only the low frequency subgraphs were taken,and then the Zernike moments of the low frequency subgraph were calculated and used as the original image features.The classification was performed according to the feature.The experimental results show that the proposed method has a relatively small amount of computation,good noise immunity and higher recognition rate compared with the method of computing the Zernike moment of the original image as the feature.At the same time,compared with the newly developed convolutional neural network recognition method,the proposed method has almost the same recognition rate and relatively small amount of computation.
展开▼
