kNN_OCR123

MNIST Optical Character Recognition

 /$$       /$$   /$$ /$$   /$$        /$$$$$$                       /$$    /$$$$$$   /$$$$$$ 
| $$      | $$$ | $$| $$$ | $$       /$$__  $$                    /$$$$   /$$__  $$ /$$__  $$
| $$   /$$| $$$$| $$| $$$$| $$      | $$  \ $$  /$$$$$$$  /$$$$$$|_  $$  |__/  \ $$|__/  \ $$
| $$  /$$/| $$ $$ $$| $$ $$ $$      | $$  | $$ /$$_____/ /$$__  $$ | $$    /$$$$$$/   /$$$$$/
| $$$$$$/ | $$  $$$$| $$  $$$$      | $$  | $$| $$      | $$  \__/ | $$   /$$____/   |___  $$
| $$_  $$ | $$\  $$$| $$\  $$$      | $$  | $$| $$      | $$       | $$  | $$       /$$  \ $$
| $$ \  $$| $$ \  $$| $$ \  $$      |  $$$$$$/|  $$$$$$$| $$      /$$$$$$| $$$$$$$$|  $$$$$$/
|__/  \__/|__/  \__/|__/  \__//$$$$$$\______/  \_______/|__/     |______/|________/ \______/ 
                             |______/                                                        

Installation

Download the project, and in the folder src, run

touch .depend
make depend
make
make cleanup

Usage

$./kNN_Ocr123 -h
Usage : ./kNN_Ocr123 [-h] [-v] [-t] [-p INDEX] [-u] [-d DIST] [-kd] TRAIN_NB TEST_NB K

Recognize images from the MNIST data base.

Positional arguments :
    TRAIN_NB                  The number of train images
    TEST_NB                   The number of test images
    K                         The kNN parameter

Optional arguments :
    -h, --help                Print this help message and exit
    -v, --verbose             Show confusion matrix and elapsed time
    -t, --test                Run tests and exit (ignore positional arguments)
    -kd, --kd-tree            Use a kd tree
    -u, --unpad               Preprocess the image by removing the padding
    -d DIST, --distance DIST  Give the used distance. Possible values are :
                                  - 0 : the square of the euclidean distance (default) ;
                                  - 1 : same, but only in the 20 x 20 pixels
                                        center of the image ;
                                  - 2 : binarize image and use Jaccard's distance.
    -p INDEX, --print INDEX   Print the image at position INDEX and exit (ignore
                              positional arguments)