基于格式塔心理学的手写识别及MatLab仿真

Handwriting recognition based on Gestalt psychology and MatLab simulation

文章从格式塔心理学的角度提出了核函数的设计方法。该方法强调:数字不等于像素点的组合,主张从整体出发,理解每个像素部分。依据这个原理,本文对图像及像素进行了两项有创新的抽象和假设。此外,由于使用了独创的不对称流动核函数和"二次学习"算法并借鉴了图论的思想,此核有很强的适应能力,能自动调整自身以适应:识别阶段(粗识别和细识别阶段)、图像内容(如:识别1和9时核的形态不同)和图像质量(数字扭曲变形程度)。以上这些,解决了传统核方法在参数调整和核函数选择上的一些困难。本文从信息论、反馈和经验校正系数的角度分析了该核的优势,并使用MatLab对所提出的算法进行了测试。结果表明,识别率比普通方法有显著提高。

According to Gestalt psychology, a novel way is proposed to design kernel function. It is emphasized by the method： the whole image of number is not the same as the combination of pixels and thus every part should be comprehended in view of the whole picture. Based on this theory, two innovative abstraction and hypothesis for images and pixels were suggested. In addition, the unique use of asymmetric and elastic kernel function as well as secondary learning. algorithm and graph theory makes kernel highly adaptable and be able to automatically adjust itself to identifying stage, image content and the image quality. All of these deal with the problems faced by traditional kernel method. Moreover, advantages of new kernel from information theory, feedback and experiential correction factor were analyzed. MatLab was used to test the new algorithm. The results proves that the recognition rate increase significantly.