Following up on my article from yesterday, here’s a full color version that can classify images using almost entirely vectorized processes, and the runtimes are excellent, though I’m still testing it, and going to apply it to a wider set of datasets –

As applied to a simple dataset I made up of three grocery items, a lime, an avocado, and a grapefruit (30 images total, 10 per class), it takes about 1/10 of one second to identify an item with perfect accuracy, which certainly makes it usable as a scanner at an actual grocery, though again, I’m going to test it more, and write a more formal article on the technique.

Below are three sample images, together with representative super-pixel images actually fed to the classifier algorithm.

Here’s the command line code:

Here’s the code that displays the super-pixel images:

Here’s the full image dataset.