Projects for Video Series titled "Practical OpenCV3 Image Processing with Python"
In this volume we will start with applying image processing techniques to various sources and try to built an intuition for how image processing is an essential step in every large scale computer vision pipeline. We will learn about what good features are in this Volume and methods to extract them. We will be focus on learning algorithms that can train over these features in the Volume 2 with applications being applying various filters, performing edge detection, thresholding, image processing and transformation techniques, working with blobs/contours and image segmentation techniques. The next volume will deal with application that combine Computer Vision techniques with machine learning to build more intelligent application like real-time face tracking, head pose estimation , gesture tracking using 2D cameras etc.
Understanding Volume 1 is crucial to get started with the Volume 2 , so it is expected to complete the entire Volume 1 along with the applications based on it.
In order to run the examples you will need to setup OpenCV3 + Python3 + TensorFlow The codes in this repository has been tested on MacOSX as well as Ubuntu 16.04.
To make it easier to setup the environment hassle-free, simply build the docker image to run and experiments with the notebook and the projects.
-
Install Docker for the targer Host(If not already installed) https://www.docker.com/get-docker
-
Cloning the Project Repo and Build/Run a Docker image to run the examples, run:
git clone git@github.com:riaz/Practical_OpenCV3_Python.git cd Practical_OpenCV3_Python docker build -t pop . docker run -it --rm -p 8888:8888 -v `pwd`:/src pop /bin/bash
Lets now see the contents of each section
In this section, we learn about the various image transformation techniques like Hough Transformation , which are all based on scoring probabilities of existence of points of interests and converging on the output. We will learn techniques to stretch, shrink, rotate and warp and image and in later sections we will see how such transformations don’t effect when trying to do Object Recognition using homography. We next learn about how Image Histograms are built and how we can use techniques like Histogram Equalization to De-noise a image effectively and we will further delve into properties of a histogram and how it can be used to build a image search search of some point of accuracy.
Query Image
Result
In this section, we learn about the Image Segmentation methods and methods to extract region of interests (ROIs) or contours on which we can apply any type of image processing pipeline to work with the contours. We also learn a technique called as template matching which can be used to detect a pattern a an image in a linear way. We also learn about Background Subtraction, which can be useful to segment away foreground from background and manipulate them individually. We also learn techniques. We will also learn about how Computer Vision is used in the field of Medical Imaging and we conclude this section by learning how to train a application to be able to detect predefined targets and also to be able to detect number plates, even though we would dive into the details of how the SVM implementation works to detect number plates. Number Plate Segmentation
Number Plate Extraction
Detection and Prediction
In this section, we learn about what features means in terms of OpenCV and what are the elements of good features in an image which may include edges, corners etc. We later explore on the most common corner detection algorithm which is Harris Corner Detection Algorithm. We also learn about SIFT,SURF et al, which are scale and rotation invariant corner detections and have application in object tracking. We then learn about optical flow which is the pattern of apparent motion of image objects between two consecutive frames caused by the movement of object or camera.We will also diving into the application of Deep-Learning for Feature Extraction on a greater scale of accuracy.
This Section has a very challenging project where we try to write deep-learning algorithms to understand scene’s and label objects and classify them accordingly. We could further extend this concept by paraphrasing the objects and their actions and coming up with a beautiful prose that summarises these elements in a image and is converse to a Text-to- Imagery Storytelling which is very popular nowadays especially in VR world.
