Relief算法在笔迹识别中的应用

Relief及其扩展算法是基于最大化假设间隔的特征选择算法，能够快速进行高维度的特征选择。该文围绕汉字笔迹识别，探讨了多类别、样本数量偏差情况对算法过程的影响。文中提出了一种有效应对数量偏差的算法过程，并成功运用于约简高维的笔迹特征。实验表明，改进后的算法不仅节约了处理时间，也进一步改进了特征选择的有效性。

Application of Relief in handwriting recognition

Relief and its extensions are feature selection algorithms based on the maximum hypothesis margin principle. They can reduce high dimensionality feature rapidly. Fouce on the handwriting identification task, how the multi-class and unbalance data affect the algorithm process was studied and a new algorithm was given. By assigning the parameters related to the number of examples, the method was appiled in reducing the high dimensionality handwriting features. Experiments indicate that this method not only saved computing time, but also resulted in a substantial improvement in the feature selection.