Packt+ | Advance your knowledge in tech

You're reading from Apache Spark Deep Learning Cookbook Over 80 best practice recipes for the distributed training and deployment of neural networks using Keras and TensorFlow

Product type Paperback

Published in Jul 2018

Publisher Packt

ISBN-13 9781788474221

Length 474 pages

Edition 1st Edition

Languages

Scala

Tools

Apache Spark

Concepts

Deep Learning

Authors (2):

Ahmed Sherif

Amrith Ravindra

View More author details

Table of Contents (15) Chapters

Preface

1. Setting Up Spark for Deep Learning Development FREE CHAPTER

2. Creating a Neural Network in Spark

3. Pain Points of Convolutional Neural Networks

4. Pain Points of Recurrent Neural Networks

5. Predicting Fire Department Calls with Spark ML

6. Using LSTMs in Generative Networks

7. Natural Language Processing with TF-IDF

8. Real Estate Value Prediction Using XGBoost

9. Predicting Apple Stock Market Cost with LSTM

10. Face Recognition Using Deep Convolutional Networks

11. Creating and Visualizing Word Vectors Using Word2Vec

12. Creating a Movie Recommendation Engine with Keras

13. Image Classification with TensorFlow on Spark

14. Other Books You May Enjoy

Leave a review - let other readers know what you think

Preprocessing images

In the previous section, you may have noticed how all the images are not a front view of the face profiles, and that there are also slightly rotated side profiles. You may also have noticed some unnecessary background areas in each image that needs to be omitted. This section will describe how to preprocess and handle the images so that they are ready to be fed into the network for training.

Getting ready

Consider the following:

A lot of algorithms are devised to crop the significant part of an image; for example, SIFT, LBP, Haar-cascade filter, and so on.
We will, however, tackle this problem with a very simplistic naïve code to crop the facial portion from the image. This is one of the novelties of this algorithm.
We have found that the pixel intensity of the unnecessary background part is 28.
Remember that each image is a three-channel matrix of 200 x 200-pixels. This means that every image contains three matrices or Tensors of red, green, and blue pixels with an intensity...