基于非下采样shearlet梯度方向直方图和稀疏表示的脱机手写数字识别.doc

基于非下采样shearlet梯度方向直方图和稀疏表示的脱机手写数字识别 基于非下采样Shearlet梯度方向直方图和稀疏表示的脱机手写数字识别 摘要:为了更有效的提高脱机手写体数字识别的性能和识别率，提出基于非下采样 shearlet梯度方向直方图特征和稀疏表示的脱机手写字符识别方法。首先对字符图像进行非 下采样Shearlet变换，得到低频子带图像和高频子带图像，然后将子图划分为若干矩形子块， 分别统计子块区域的梯度方向直方图分布，将分块直方图串接起来形成非下采样梯度方向直 方图特征(HNSCOG)，最后用HNSCOH特征构建超完备字典，通过稀疏表示重构最小误 差实现字符图像分类。在MNIST和USPS数据集上测试，与DDH+SVM方法、Sparse LS-SVM 方法、sub-sampling+SVM方法和MTC+linear-SVM方法的识别率比较，实验结果表明， HNSCOG和稀疏表示的方法可以较大地提高脱机手写数字的识别率。 关键词:脱机手写数字识别，非下采样Shearlet变换，梯度方向直方图，稀疏表示 Offline handwritten digit character recognition based on histograms of nonsubsampled shearlet oriented gradients and sparse representation bstract: In order to improve recognition rate of handwritten digit character efficiently, a novel A method of handwritten numerals character recognition based on histograms of nonsubsampled shearlet oriented gradients features (HNSCOG) and sparse representation was presented in this paper. Firstly, the nonsubsampled shearlet transform (NSST) was utilized to decompose the cha-racter images on various scales and in different directions, and the low frequency sub-band and bandpass sub-band coefficients were obtained. Then, sub-images of digital character was divided into grids of blocks and cells to extract HOG features, and the histogram of each unit was com-puted and concatenated as HNSCOG features descriptor. Finally, HNSCOG features were com-bined to form an over-complete dictionary which was employed by sparse representation to clas-sify the handwritten numerals character. To compare the performance of the proposed method, several alternative algorithms for handwritten number recognition had been considered, for in-stance distance distribution histogram(DDH) plus SVM method, Sparse LS-SVM method, sub-sampling plus SVM method and MTC plus linear-SVM method. In order to evaluate these techniques, a collection of well known standard databases had been used: MNIST digit dataset and USPS digit dataset. The experimental results indicate that the handwritten numeral charact