Convolutional Neural Networks — Image Classification w. Keras

There’s no shortage of smartphone apps today that perform some sort of Computer Vision task. Computer Vision is a domain of Deep Learning that centers on the fundamental problem in training a computer to see as a human does.

The way in which we perceive the world is not an easy feat to replicate in just a few lines of code. We are constantly recognizing, segmenting, and inferring objects and faces that pass our vision. Subconsciously taking in information, the human eye is a marvel in itself. Computer Vision deals in studying the phenomenon of human vision and perception by tackling several ‘tasks’, to name just a few:

Object Detection
Image Classification
Image Reconstruction
Face Recognition
Semantic Segmentation

The research behind these tasks is growing at an exponential rate, given our digital age. The accessibility of high-resolution imagery through smartphones is unprecedented, and what better way to leverage this surplus of data than by studying it in the context of Deep Learning.

In this article, we will tackle one of the Computer Vision tasks mentioned above, Image Classification.

Image Classification attempts to connect an image to a set of class labels. It is a supervised learning problem, wherein a set of pre-labeled training data is fed to a machine learning algorithm. This algorithm attempts| to learn the visual features contained in the training images associated with each label, and classify unlabelled images accordingly. It is a very popular task that we will be exploring today using the Keras Open-Source Library for Deep Learning.

The first half of this article is dedicated to understanding how Convolutional Neural Networks are constructed, and the second half dives into the creation of a CNN in Keras to predict different kinds of food images. Click here to skip to Keras implementation.

Let’s get started!