A little background:
I work for an image processing company that creates inspection systems and sell it as a product. These systems are used to check the quality of prints. When we say "check" it looks for defects like dirt, streak, wrinkles, missing characters, and other foreign materials that should not be part of the printed product. We use computer vision to connect either camera(s) or flat-bed scanner to capture images. We then perform processing in these images to decide whether it is has defect or not. If we need speed, all the processing goes into the hardware, otherwise other processing goes to the software side.
All of this processing requires setting of parameters (a master image as point of reference, inspection levels, sizes, misalignment tolerance) for the system to perform well in doing its tasks. Is it possible that the system automatically learn what right parameters to use to accurately detect these defects?
All detected defects when categorized are classified via its composition, width, height, but is there a better way to classify these defects by just learning from it's previous data? Meaning, if a user selected few defects and identified that group as "wrinkle", can the system learn from it and automatically classify these defects with higher accuracy?
This is where my journey starts to learn AI (Artificial Intelligence) and it's subset ML (Machine Learning). Although I've been working with computer vision industry, being able to analyze image and getting results from it, I feel that the intelligence it deserve is still lacking. Is AI/ML the key? We'll see... I've been fascinated with the science behind AI so maybe I'll find the answers (or maybe not). For now I want to learn more on how it works so I can design and develop effective solutions in this era of artificial intelligence and machine learning.
Since I graduated wayback 2001, I'm a little bit rusty (or rather beginner) on other required areas. So I feel like the bottom-up approach in this self-paced track I designed is the best one to follow: