统计不相关性的最佳鉴别变换的手写数字识别

Handwritten digit recognition based on statistical unrelated discriminant transform

为了进一步深入研究推广手写体数字识别技术,介绍并验证了具有统计不相关性的最佳鉴别变换在手写体数字识别中的优越性。与经典的Foley-Sammon鉴别变换法相比,具有统计不相关性的最佳鉴别变换相关性更小,提取的特征更有效。提出利用具有统计不相关性的最佳鉴别变换来提取特征并结合BP网络设计分类器用以实现手写体数字识别。通过3个对比实验证实了基于具有统计不相关性的最佳鉴别变换方法的识别方法的有效性。

To promote the research and application of handwritten digital recognition technology, the superiority of statistical unrelated optimal discriminant transform in handwritten digit recognition was presented and certified. Compared with classic Foley-Sammon discriminant transform, statistical unrelated optimal discriminant transform extracts features with less relativity. Statistical unrelated optimal discriminant transform was proposed to extract features and BP network to design sorter, a way to recognize handwritten numeral. Finally, the validity of the statistical unrelated optimal discriminant transform was testified by three comparative experiments.