Machine Learning for OpenCV

Understanding the kernel trick

Granted, we won't have time to develop all the mathematics needed to truly understand the kernel trick. A more realistic section title would have been Acknowledging that something called the kernel trick exists and accepting that it works, but that would have been a bit wordy.

Here's the kernel trick in a nutshell.

In order to figure out the slope and orientation of the decision hyperplane in the high-dimensional space, we have to multiply all the feature values with appropriate weight values, and sum them all up. The more dimensions our feature space has, the more work we have to do.

However, mathematicians smarter than us have long realized that an SVM has no need to explicitly work in the higher-dimensional ...