Multi-Scale Oriented Basic Image Features Column for Handwritten Digit Recognition

Abstract

Handwritten digits recognition is a key research problem in the domain of image analysis and pattern recognition. Specifically, the appearance approaches based on feature extraction have been proposed to solve many research issues. This paper presents a novel way of extending the oriented Basic Image Features column (oBIFs column) to multi-scale features. Moreover, this method is very efficient and robust for hand handwritten digit recognition as the changes in size, slant and shape have little effect on the accuracy. The proposed method is carried in two steps: first, we concatenate oBIF image for scale parameter at two or more scales to provide Multi-scale oBIFs. Second, the two variation Multi-Scale oBIFs are crossed to form columns at each location. Finally, a histogram is created from the occurrences of different patterns. The CVL dataset was considered to evaluate our approach, where the digits used where from different fonts, widths, and directions. We managed to achieve a good recognition of unnormalized digits in almost 96% of the cases, which is comparable with the state-of-the-art approaches. © 2021, The Author(s), under exclusive license to Springer Nature Switzerland AG.