Pattern Storage & Recalling Using Hopfield Neural Network and HOG Feature Based SVM Classifier: An Experiment with Handwritten Odia Numerals
Naman Sharma, Karan Kalra, Pradeepta Kumar Sarangi, Lekha Rani, Merry Saxena, Santosh Kumar · 2023
Handwritten Odia text identification is very useful for the society and also a challenging task as it may include a variety of writing styles, complicated character-touching, and abundant categories of character class. This work implements two different classifiers for handwritten Odia numerals. The first experiment involves the implementation of a Hopfield neural network with binary image as the feature set. Using the feature extraction technique based on the Histogram of Oriented Gradients (HOG), a Support Vector Machine (SVM) classifier is implemented in a second experiment. Different cell sizes have been used to extract HOG features and finally the best cell size with optimal features have been used. The data set has been created on Microsoft Paint by drawing the digits manually and also collected from different users. The Hopfield model has been implemented with four different data sets (0% noise, 10% noise, 20% noise and 30% noise). Addition of noise to the data has been done through programming. Both the models (Hop field and SVM) have been implemented through MATLAB and the data set has been used for testing the models. Results prove that the Hopfield model's performance is higher than the SVM classifier whereas, the recognition accuracy for Hopfield model is decreasing with increase in noise percentage and could be considered as poor performance.