New Approach of Features Extraction for Numeral Recognition

Abstract

This paper proposes a new approach of features extraction based on structural and statistical techniques for handwritten, printed and isolated numeral recognition. The structural technique is inspired from the Freeman code, it consists first of contour detection and closing it by morphological operators. After that, the Freeman code was applied by extending its directions to 24-connectivity instead of 8-connectivity. Then, this technique is combined with the statistical method profile projection to determine the attribute vector of the particular numeral. Numeral recognition is carried out in this work through k-nearest neighbors and fuzzy min-max classification. The recognition rate obtained by the proposed system is improved indicating that the numeral extracted features contain more details.

Multiple Classifiers and Invariant Features Extraction for Digit Recognition

“…A new method is proposed by [16], of handwritten pre-processing and printed and isolated numeral characters. Were the size of the input image of each numeral was reduced by discarding the redundant information.…”

Digit recognition using multiple classifiers

“…A feature vector is the number of on pixels in each zone (The zoning vector should be normalized by dividing by the total surface of the zone) [2] [4]. This operation reduces widely the size of the image and preserves the useful information for recognition of every input image [6]. In our case we used a zoning 2x2, 2x1, and 1x2.…”

“…A new approach is proposed in [6] of preprocessing for handwritten, printed and isolated numeral characters. The new approach reduces the size of the input image of each numeral by discarding the redundant information.…”