Image Classification

What is image classification?

The process of taking an image or picture and getting a computer to automatically classify it, or try providing the probability of the class of the image

*A class represents a label for instance: cat, dog, house, etc*

Image Classification has many popular use cases and is used in everyday life!

Examples of using image classification

-- used in radiology to help medical professionals identifying anomalies in X-ray

-- used in self driving cars to classify images around them to help the car identifying how to navigate the road

Targets: usually start with a subset of categories or closes

ex) cat and dog

y=0 refers to cat/y=1 refers to dog

Computers can't understand the image of dog and cat but they can understand the intensity values of a digital image

We will use the intensity values to classify the image

We let X represent the image, we want to classify

In the RGB case, it's a three dimension array or tensor

**All images must have the same number off rows and columns

The data set is a set of images X and labels Y

The comma is used to indicate that they are together

ex) X4,Y4 = 1 refers to dog

MNIST database, a large database of handwritten digits

Y can take on the value from 0 to 9

Here are 160 of the 60,000 small square 28 by 28 pixel grayscale images of handwritten style digits

The final goal of the module is to come up with python function that takes the input image and outputs a class

Challenges of Image Classification

There are many challenges with the image classification

Consider the cat and dog example

Change in viewpoint

change of illumination/deformation/occlusion/background clutter

Some several machine learning supervised methods for image classification

1. K-- nearest neighbors

2. Feature extraction

3. Linear classification

Copyright Coursera All rights reserved