Handwritten Digit Recognition Method Based on DTW and SVM

Handwritten digit recognition is a classic and fundamental problem in the field of machine learning. In recent years, with the development of deep learning theory, deep neural networks have become the main solution to this problem, achieving higher recognition accuracy. However, there are still challenges related to interpretability, lack of theoretical guidance in network structure design, overfitting, and high reliance on data. In this paper, we propose a method for handwritten digit recognition based on the combination of Dynamic Time Warping (DTW) and Support Vector Machine (SVM). Firstly, we utilize the Radon transform to convert the two-dimensional images into multiple one-dimensional data, achieving rotational invariance of the images. Then, we apply DTW technique to align the signals, making the recognition algorithm insensitive to local and global deformations of the images. Based on this, we construct a Gaussian kernel function by considering both alignment distance and deformation distance of the signals. Finally, we employ the well-established SVM technique to accomplish the task of handwritten digit recognition. Experimental results demonstrate that our proposed algorithm achieves state-of-the-art recognition rate, with better generalization performance, validating the effectiveness of our method.