Face Detection in Python Using OpenCV

What is OpenCV?

OpenCV (Open Source Computer Vision Library) is an open-source computer vision and machine learning library.
It is a BSD-licensed product, free for both business and academic purposes.

The library provides more than 2500 algorithms, including:

• Machine learning tools for classification and clustering
• Image processing and vision algorithms
• Basic algorithms and drawing functions
• GUI and I/O functions for images and videos

Applications of OpenCV

Some common applications of OpenCV include:

• Face detection
• Object recognition
• Extracting 3D models
• Camera calibration
• Image processing
• Motion analysis

OpenCV is written in C/C++ but has interfaces for C++, C, Python, and Java. It supports Windows, Linux, Mac OS, iOS, and Android. It is optimized for computational efficiency and designed for real-time applications, leveraging multi-core processing.

---

Face Detection

Face detection has gained significant attention due to its real-time applications.
However, it is a challenging task for machines because of variations such as:

• Pose differences (front, side, angled views)
• Occlusion (partially hidden faces)
• Image orientation
• Lighting/illumination changes
• Facial expressions

OpenCV provides many pre-trained classifiers for face, eyes, and smile detection.
For face detection specifically, two classifiers are commonly used:

1. Haar Cascade Classifier
2. LBP Cascade Classifier

We will explore both in this tutorial.

---

Haar Cascade Classifier

The Haar Cascade Classifier is a machine-learning–based approach proposed by Paul Viola and Michael Jones.
It uses a cascade function trained on positive images (with faces) and negative images (without faces).

Algorithm Stages

1. Haar Feature Selection

   • Haar features are computed on subsections of the image.
   • They calculate differences in pixel intensity between adjacent regions.
   • A large number of Haar-like features are required for facial recognition.

2. Creating an Integral Image

   • To speed up calculations, an integral image is used.
   • This reduces computation to only four pixels instead of all.

3. Adaboost

   • Not all computed features are relevant.
   • Adaboost selects the most important features for classification.

4. Cascading Classifiers

   • Features are grouped into stages.
   • Each stage filters out non-facial regions.
   • Only regions passing all stages are classified as faces.

---

LBP Cascade Classifier

The Local Binary Pattern (LBP) classifier is based on texture descriptors.
Since the human face is composed of micro-texture patterns, LBP features help distinguish faces from non-faces.

Algorithm Steps

1. LBP Labelling

   • Each pixel is assigned a binary label.

2. Feature Vector

   • The image is divided into sub-regions.
   • A histogram of labels is created for each sub-region.
   • These histograms are concatenated into one large feature vector.

3. Adaboost Learning

   • Gentle Adaboost is applied to remove redundant information.
   • A strong classifier is built from useful features.

4. Cascade of Classifiers

   • Classifiers are arranged in a cascade.
   • Each stage filters sub-regions of the image.
   • Only facial regions survive all stages.

---

Steps for Face Detection

1. Load the Haar Cascade Face Algorithm
2. Initialize the Camera
3. Read a Frame from the Camera
4. Convert the frame to Grayscale
5. Obtain Face Coordinates using the classifier
6. Draw a Rectangle around detected faces
7. Display the Output Frame

---

Output

The detected face(s) will be highlighted in the video feed.
To view the output, check the generated media file or run the script live.