Digit-Recognizer-via-Metal

There is an application for course project in High School of Economics, Software Engineering department. It is backed on the Metal framework for handwritten digits recognition.

Key points:

• Firstly, that application relies on CNN, which implemented by Graph API, introduced in Metal 2.
• Secondly, for connected compoment labeling task, I am using disjoint set uniond data structure and 2 path linear algorithm.
• Finally, before putting an image to the CNN, it has to be prepared. For that purpose there is an application of four filters: Gaussian blur, Image binarization, Morphological erosion and dilation.

Sample:

Input - raw data from back camera.

Output - binarized image with detected and recognized digits.