OpenCV Advanced Features

Deep learning is a subfield of machine learning, based on traditional neural networks and convolutional neural networks. It's state-of-the-art pattern recognition. Deep learning is reaching the best accuracy in fields such as speech recognition, text recognition, and image classification. Actually, OpenCV adds the deep learning module as a basic module in its core algorithms and is making big efforts to increase its performance using CPUs and GPUs.

In this chapter of advanced features, we are going to cover how to enable Halide, a new language that OpenCV is using to improve the performance of deep learning inferences.

Finally, we are going to explore the new possibilities of using OpenCV in a web browser as a JavaScript library, which allows us to use computer vision in a web client without installing anything on the client computer.

In this chapter...

Face detection using deep learning

In this recipe, we are going to learn how to use a trained deep learning model for a face detection algorithm in OpenCV. To do this, we are going to download pre-trained face detection models and use OpenCV methods to import the model and also see how to convert an input image or frame into the required deep learning structure.

How to do it...

Using deep learning in OpenCV is very easy, and the only files that we require are the pre-trained models and know the basic configuration of it. We can download tested, pre-trained OpenCV models from https://github.com/opencv/open_model_zoo.

To create the face detector algorithm, follow these steps:

1. Download and save the model of the face detector...

Object detection with YOLOv3

In this recipe, we are going to use the YOLOv3 DNN algorithm, which performs object detection. Object detection in an image is a common task in computer vision; with deep learning techniques, we can achieve highly accurate detections. YOLOv3 reaches a 60.6 mAP in the COCO dataset (80 classes and more than 300,000 images), which has a very good precision with 20 fps or 33 mAP with 220 fps.

How to do it...

In this recipe, we are going to use the same functions and classes to load the model, convert the images, and predict, but we are going to introduce one more important function – non-maximum suppression – and a few util functions to draw the predictions with their labels:

1. Create...

Enabling Halide to improve efficiency

Halide is a new programming language written for high-performance image processing on new computers. Halide currently targets the following:

• CPU architectures: X86, ARM, MIPS, Hexagon, and PowerPC
• Operating systems: Linux, Windows, macOS, Android, iOS, and Qualcomm QuRT
• GPU Compute APIs: CUDA, OpenCL, OpenGL, OpenGL Compute Shaders, Apple Metal, and Microsoft Direct X 12

In this recipe, we are going to show how to compile OpenCV with Halide and how to enable it in your DNN algorithms.

How to do it...

To enable Halide in OpenCV, follow these steps. It will take a while to compile it all:

1. Download and unpack the LLVM source code from http://releases.llvm.org/4.0.0/llvm-4.0.0.src.tar.xz...

OpenCV.js introduction

This recipe introduces a new way to develop computer vision algorithms with OpenCV using JavaScript for the World Wide Web (WWW). Before OpenCV.js, developers that needed some computer vision tasks on the web had to develop them in C++ in servers or Java applets clients, which were not optimized, but with OpenCV.js, the new paradigm of client applications using web browsers has grown with unlimited of possibilities.

In this recipe, we are going to create a very simple case where we load an image into an img HTML tag, change the color image, and present it to a canvas element.

How to do it...

Before starting to use OpenCV.js, we need a web page as a user interface. To get a user interface, we are going...