Steady Model for Classification of Handwritten Digit Recognition

Handwritten digit recognition plays an important role not only in computer vision but also in pattern recognition. Handwritten digit recognition is the competence of a machine to receive, calculate and decipher a human handwritten input from sources such as handwritten manuscripts, especially created before the advent of a digital revolution and digital images.This work implements the system to read the handwritten digits with a custom novel method identical to the amalgamation of different techniques, including principal component analysis, support vector machine and K-nearest neighbours to recognize and classify handwritten digits into their respective labels. PCA algorithm finds out the best linear combinations of the original features so that the variance along the new feature is maximum. Recognition of characters is done using KNN nonparametric machine learning algorithm, and SVM lowers the generalization error of the overall classifier. The proposed work does the analysis on digit data set having a total of 70,000 image samples. The performance of the system is analysed using different measurement metrics like precision, recall, f1 score and support, and the recognition of the patterns in the images shows the result with classification accuracy of 97%.