Pattern Recognition

Pattern Recognition

Pattern recognition is an essential aspect of analysis. Probably most of the analysis that we do today is related to patterns. Humans know that a thing (an abstract object) is ball because that “thing” matches the pattern of a ball (it’s roundness, it’s size, it’s texture and etc). We know that the “thing” that seats next to you is another human being because it matches the pattern of a human (the built, the face, the feature, the texture, the smell, the sound, and etc).

When humans look at a “thing”, information from different senses are extracted and that information is feed to the brain and then “probably” the brain processes it and compare it with the existing patterns it met before. If the “sample image” is not in the library of patterns, it registers it as something new. The brain then tells the rest of the body that the “sample image” is something new. If the brain recognizes the pattern it then looks at the behavior tagged to that pattern. It “probably” asks “what does this pattern do?” It then “recalls” the past behavior of the pattern and from there analyzes the future behavior of that pattern.

An example scenario would be, say you are watching a ball game, and you are enjoying the game then suddenly the ball flew up in the air, right above you. At that moment, your brain took a sample of the event. It then realized that the thing flying over you is a ball and from the past experience or from an experienced that you witnessed you know that what goes up usually goes down. It also then realized that the ball should fall on you or near you. It also realized that the ball (being a bowling ball) is hard and it should be painful when it hits your face. Your brain signals the danger alarm and it will cause you to run for your life.

Today, the in the computer science world, we can build a simple A.I. (Artificial Intelligence) that will analyze simple events. What it will do is to mimic what the brain does basing on what is discussed above. To analyze we need to:
1. Capture the event. (Actually just get a sample image)
2. Generate a model of the event. (Realize that the event here could be a thing)
3. Check if the model is already recorded in the “Event’s Library”.
4. If it’s not yet there, saved the event.
5. If it’s there don’t record.
6. Extract the behavior tagged to that event.
7. There should also be a way to get another sample of the event and tagged the recorded attributes to the existing tags.


That should make our lives less complicated. (You wish.) Seven steps that looks simple but honestly, these steps lacks a lot of things. There are more to do than this and each step has more than 5 sub-steps. But I think this can be done, we just need to make our requirements as simple as it can be.

We need to limit our requirements otherwise we will get nowhere. For simplicity, we just need to spot a 2-dimensional “thing” and predict its behavior in a 2-dimensional space.

Speaking of 2-dimensional space, another aspect of analysis via pattern recognition is what I call “spatial awareness”. The observer should be aware what space the object is moving. So in this case, we don’t want to care bout that since we assumed that we are observing a 2-dimensional object in a 2-dimensional space.

What we need to do now is to capture the “image” of a 2D object. It’s height and width and other information necessary to describe the physical appearance of the object (including the color). Record it and place it in the library. Get a sample of the behavior of the object and then record it.

// To be continued….