Handwritten digit recognition using a column scheme-based local directional number pattern

Abstract

One of the most well-known challenges in computer vision and machine learning is the recognition of handwritten digits. This study presents an advanced approach to improving isolated-digit recognition through the use of advanced feature extraction techniques. For example, digit recognition is commonly used to read numbers on forms and checks in banks. This paper introduces a novel method of extending the local directional number pattern (LDNP) to a column scheme using two different masks and their resolutions. A new descriptor of the LDNP column scheme is being proposed that combines derivative Gaussian and Kirsch masks in order to enhance textural analysis and capture more detailed local textual information. This approach is highly efficient and robust, able to handle variations in size, shape, and slant. Additionally, the support vector machine (SVM) is employed as a classifier, which has been shown to make better decisions. The empirical investigation is carried out using the CVL dataset, resulting in recognition rates that are comparable with the latest advancements in the field. The overall precision of 96.64% is achieved, outperforming existing similar works.