Getting started with image classification

Key Takeaways

[Music] the concept of having a computer recognize the contents of an image and not just see the data making up the image is called computer vision it comes in many flavors with the most common being image classification or labeling where the primary content of the picture is determined or there's object detection where common objects in the image are found and boxes highlighting where they're found in the image are provided in this learning path we're going to take a look at image classification where you give a computer an image and it will tell you what it thinks that image contains before you get down to coding let's explore how this works from a machine learning perspective a neural network generally looks something like this you have a number of layers and every neural network will have an input layer an output layer and a number of hidden layers i won't go into detail on how these work in this video but if you check out the machine learning foundation's free course on youtube you'll see how to build models with all of these layers as well as things like convolutional neural networks to perform feature extraction and this allows you to build very smart image classifiers but the important thing for us to consider for now is the architecture of a neural network's output layer for example in this simple diagram the output layer has two neurons and this means it can be used to recognize only two classes of image for example a dog or a cat and when you feed an image into the neural network the output will give you the probabilities of each of these classes thus if you're using a neural network model in your apps you'll have a black box that you feed in an image and you'll get numbers out of it and it's up to you to parse through these numbers to determine what the image is in this learning path you'll start with ml kit which gives you a built-in image labeler that recognizes over 600 classes of image giving you probabilities for each so for example if you feed it with a picture of a cat it might return high probabilities for cat pet animal but low ones for basketball car or truck or as we see when we'll go through the code lab if you give it a picture of a flower it might give you a classification of plants and petal but not a specific type of flower like a daisy and that's something that you're going to see how to do later in this series you'll learn how to create a simple app that recognizes over 600 types of image and let's take a quick look at what that looks like on both android and ios so here's the app running in android studio i can click the label image button and i'll get a bunch of labels back like petal flower or plants now this works by getting the picture as a bitmap and then creating a labeler object and an input image to it from my bitmap it will then process the image with the labeler and when it's done i'll get a set of labels back and these contain the confidence values now that's the output from the neurons with probability that we showed earlier on and we can just show these in our app if we wanted to on ios it's very similar here's the app running in xcode and when i click classify i'll get back a set of labels and confidence values for each label and the code works in pretty much the same way i get a labeler object i process my image using that labeler object and then i'll process the labels that come back from it and then i can just iterate through them get the confidence values and write those to the screen and you can see the results in the view here okay now that you've seen this in action it's time to get coding in the next code lab you'll see how to build this app on both android and ios happy coding [Music] you