FULL LENGTH ARTICLE OPEN ACCESS Recognition of Persian Handwritten Numbers using LBP - HOG Desc riptor
Reza Talebian, Mojtaba Mohammadpoor · 2015
Recognition of handwritten numbers in any language is one of the most important issues attracted many researchers and this is because of the increasing importance of the issue in today's world and in applications such as automatic detection of mail addresses or identification of numbers of bank checks. Given the great interclass diversities and much extra - class similarities between some Persian handwritten numbers, solving the Persian handwritten numbers recognition problem is difficult. In this paper, a new method for recognizing Persian handwritten numbers using a combination of HOG and LBP descriptors is provided. The proposed descriptor enjoys significant advantages of which the important one is the recording of information and features relat ing to the image (by descriptor LBP)and yet the feature extraction of the image edges (by descriptor HOG). Another advantage of the proposed descriptor is the very small length of the feature vector and the fast calculations. To evaluate the effectiveness of the proposed method the standard data base HODA, a large database of 60,000 images, were used . Experiments carried out on the database show the high effectiveness of the proposed method (with an accuracy of 99.3%) in comparison with other similar methods.