Let's learn about image classification. So the goal of this exercise is given this image of a cat, or a dog or a panda. These images we have some set of images which are pre labeled like this in this particular case the cat will be labeled as cat. Like this dog will be labeled as dog and pilot will be labeled as panda. So this set of images if we call it as a training set, and we will be learning a technique called supervised learning. So in this particular case, we will feed the machine with a data.
That data has images on the left and the labels on the right. So we will tell the machine upfront. Hey, this is the image This is a key image of a cat. This is the image. And this is a measure for dog. This is panda.
And this is a label for it right? So this is like we school teacher is teaching children, right? First you teach them that this picture is a cat. This picture is a dog and this picture is panda. And then that's how kids learn. So exactly, we are going to follow the similar mechanism, where we are going to feed this data to a computer and say, This is a supervised data, where we already labeled and this is how cat looks like.
This is our dog looks like and this is a panda looks like after that after we feed the data. Then we will also tell the machine how thick it can learn from this data. And then once the machine learns, the next task for the machine is then we will feed another set of similar animals. So these three images for example, are completely different. Images of a cat, dog and Panda and we will not give label this particular case. Right?
The label will be question mark. And it is the job of the machine to correctly label these images as cat, dog and panda. So that's the goal. That's the challenge for the machine that we are going to give in this particular course. Though one challenge here is machines don't perceive image same way as humans. So here on the left, you can see grayscale images of black and white picture of flower tree proper plant, so humans can easily see the actual objects in the picture right, you can clearly see there is a flower then there are some small flowers just blossoming up.
There are leaves in the tree, and there is a sunlight in the background, you can clearly identify these objects very clearly right? Because human eye can identify these objects easily. But on the right, if you take a look, the machines don't see the image the same way as humans do. So this let's say this example this machine is a 10 by 10 image where the width is 10 pixels and height is 10 pixel. So what machine what computer looks at is a just as an array of 10 by Ken so it has 10 cells in each row, and 10 cents in each column. Thereby it's an array that has hundred elements, and each cell represents the intensity of that particular sale and the intensity ranges from zero to 255. zero means white so as it moved closer to 250 It becomes dark.
So each of the cells represents the darkness or the lightness of the image, just the intensity of the image. And these are just the numbers as the machines just looks at the image as a matrix of numbers, that's all it looks at. He doesn't understand where exactly the flower is unless until we train them. But this is the basic challenge in machine learning for computer vision, so machines see the image just as set of numbers. Now this is this is just a grayscale image right? It becomes a little bit more challenging when you have color image.
Let's look at it next. So it gets a color image. There are multiple colors right and so this image is represented to machine a set of three different areas. So there are three primary colors right so again, we can say this is 10 by 10 image, we To 10 with a total of 100 cells 10 in each row and each column, so, this in color image is divided into three primary color channels, red channel, blue channel and green channel and each of the channel again has intensity of the cells in that particular color, right. So, color image is represented as a 10 by 10 array of three different colors. Red, blue, and green.
Right. Let's go back I would just want to make sure you understand this very clearly. So here machine see this grayscale images are just 110 by 10 array, but for the color image, the machine would see it as 310 by 10 arrays, one for each primary color point for red, one for blue and one for green. Right. So these are the ones Things that are very important things for you to understand. So if you can go back and forth and replay the video and make sure we are driving this point home because this is very fundamental requirements if you really want to understand how computers can classify images