基于旋转投影统计特征的手写数字识别方法

摘要

To extract geometrical property by abstracting the contour and skeleton of digits can obtain the particulars efficiently; but irregularity induces low recognition rate. The statistical analysis theory can overcome the weakness. First;Gap-Rate and Gap-Variation extraction methods were probed;by projecting and computing the rate and change of each gap as the eigenvector of the digit Further; Rotational Projection was probed based on the idea of enhancing the or thogonality of eigenvectors and reducing the redundancy information in the way of rotating the datum line. Theoretical analysis and experiment proved the high recognition rate of rotational projection with the same statistical properties;al so parameters were recommended in the end. Further more;it also recognizes the oblique digit.%通过抽取数字的轮廓和骨架来提取几何特征,可以有效地反映手写数字的细节,但手写数字的不规范性导致其识剐率并不高.运用统计分析理论可以克服这一缺点.首先提出了基于投影间隔比率和间隔变化的特征提取方法,通过数字投影计算间隔的像素数比率和变化趋势,并将其归一化作为特征向量.进一步通过旋转投影基准线,增加特征向量之间的正交性以减少信息冗余,基于这一思路提出旋转投影的识别方法.理论分析和实验证明了旋转投影可以在相同特征数量的情况下达到更高的识别率,并给出了推荐参数.此外,通过旋转投影,直接解决了倾斜数字的识剐问题.