TensorFlow For Dummies

Chapter 8

Classifying Images with Convolutional Neural Networks (CNNs)

IN THIS CHAPTER

Exploring image filtering and convolution

Looking at convolutional neural networks (CNNs)

Introducing the CIFAR-10 dataset

Presenting TensorFlow’s image operations

This chapter explains how you can code image recognition applications using TensorFlow and convolutional neural networks (CNNs). These applications are similar to the vanilla neural networks from Chapter 7, but they include layers specifically intended for image classification.

Filtering Images

If you’ve used image editing applications like Adobe Photoshop, you’re probably familiar with filtering tools, which add effects, such as blurring, sharpening, or embossing, to images. Mathematically, these tools perform their operations using a process called convolution. This process plays a critical role in image recognition, and while it’s not important to grasp all the gory details, it’s good to understand the general process.

Convolution

Image convolution replaces each pixel of an image with the result of a two-dimensional dot product. ...