#Abstract
The objective of the project is to study the accuracy in recognizing the hand written digits of MNIST database with statistical features using Support Vector Machine,Random Forrest,Extra-Trees Forrest machine learning algorithms.
The project contains two parts :
##1. OpenCV(SVM) + C++
It works on image database files from http://yann.lecun.com/exdb/mnist/
- Language of Implementation - C++
- Accuracy on test images using SVM from OpenCV - 85.4%
##2.Kaggle\Scikit-Learn + Python
It works on image database files from http://www.kaggle.com/c/digit-recognizer/data
- Language of Implementation - Python
- Accuracy on test images using SVM from scikit - 90.343%
- Accuracy on test images using Random Forrest from scikit - 96.971%
- Accuracy on test images using Extra Tree Forrest from scikit - 97.071%
###The features used in both parts are :
- first dark pixel in each row from left and right (28+28)
- first dark pixel in each col from top and bottom (28+28)
- first dark pixel in each row from center towards left and right (28+28)
- first dark pixel in each col from center towards top and bottom (28+28)
- number of dark pixels in each row and col in both halfs saperatly (28+28+28+28)
total features = 336