Introduction

Any facial detection and recognition program or system must have a face recognition algorithm at its core. These algorithms are divided by experts into two main categories. The geometric method concentrates on identifying features. For the purpose of extracting values from an image, photometric statistical approaches are applied. Then, in order to remove variations, these values are compared to templates. Additionally, the algorithms can be broken down into two additional groups: feature-based and holistic models. While holistic approaches look at the human face as a whole, the former focuses on facial landmarks and evaluates their spatial characteristics and link to other features.

In terms of picture recognition, artificial neural networks are the most widely used and effective technique. Numerous mathematical processes are carried out concurrently by neural networks, which are the foundation of facial recognition systems.

Three primary functions are carried out by the algorithms: identifying faces in images, videos, or real-time streams; creating a mathematical model of a face; and comparing models to training sets or databases to confirm the identity of a person.

The most well-known facial recognition algorithms and important characteristics are covered in this article. Due to the fact that each approach provides advantages that are task-specific, researchers are always experimenting with method combinations and creating new technologies.

Algorithms

1)CNN

One of the innovations in artificial neural network (ANN) and AI development is the convolutional neural network (CNN). One of the most widely used deep learning techniques teaches a model to carry out classification tasks directly on an image, video, text, or sound. In the areas of computer vision, natural language processing (NLP), and the largest image classification data set, the model exhibits outstanding results (Image Net). Convolutional and pooling layers have been added to a typical neural network to create CNN. These layers can number in the hundreds or even thousands for CNN, and each one gains the ability to recognize various image elements.

2)HAAR CASCADES

An approach to finding objects on images is called Haar Cascade. The algorithm learns from a huge number of positive and negative samples, where a positive sample contains an object of interest and a negative sample contains anything else. The classifier can identify an interesting object on fresh photos after training. Combining the technique with the local binary pattern algorithm was utilised in criminal identification to identify faces. Even with fluctuating expressions, the Haar cascade classifier requires 200 (out of 6000) characteristics to guarantee an 85-95% recognition rate.

3)Eigenfaces

The face variance in picture data sets is found using the face detection and recognition algorithm Eigenfaces. With the aid of machine learning, it encodes and decodes faces using these variations. A collection of "standardised face constituents" known as a set of eigenfaces is produced by statistically analysing many different face photos. Since this method doesn't use digital images but rather statistical databases, facial traits are given numerical values. A mixture of these variables in various percentages makes up every human face.

4)Fisherfaces

One of the most well-liked facial recognition algorithms, Fisherfaces is regarded as being superior to many of its rivals. It is frequently compared to Eigenfaces as an enhancement to the Eeigenfaces method and is seen as more effective at class distinguishing throughout the training process. The main benefit of this algorithm is its capacity to extrapolate and interpolate over variations in illumination and face expression. When used in conjunction with the PCA approach during the preprocessing stage, the Fisherfaces algorithm has been reported to have a 93% accuracy rate.

5)Kernel Methods : PCA and SVM

The principle component analysis (PCA) is an all-encompassing statistical technique with a wide range of real-world uses. PCA seeks to minimise the quantity of the source data while retaining the most crucial details when utilised in the face recognition process. It produces a number of weighted eigenvectors, which combine to form eigenfaces, which are sizable collections of various pictures of human faces. Each image in the training set is represented by a linear combination of eigenfaces. These eigenvectors are obtained from the covariance matrix of a training image set using the PCA. The primary elements of each image are calculated (from 5 to 200). Minor distinctions between faces and noise are encoded by the other components. The primary component of the unknown image is compared to the primary components of all other images as part of the recognition process.

A machine learning technique called Support Vector Machine (SVM) employs the two-group classification principle to tell faces from "not-faces" apart. An SVM model is given a labelled training data set for each category in order to classify fresh test data. For face recognition, researchers use both linear and nonlinear SVM training models. Recent findings demonstrate a greater margin and superior recognition and classification results for the nonlinear training machine.

6)THREE-DIMENSIONAL RECOGNITION

The fundamental concept behind 3D face recognition technology is the distinctive design of the human skull. The distinctive skull anatomy of each individual can be explained by a wide range of factors. This form of facial recognition works by comparing a 3D facial scan to patterns in a database. It has a crucial benefit in that the detection and recognition procedure is unaffected by makeup, facial hair, spectacles, and other such characteristics. The most recent studies have made use of a system that maps 3D geometry data onto a normal 2D grid. It exhibits the highest performance recorded on the FRGC v2 and enables the integration of the descriptiveness of 3D data with the computational efficiency of 2D data (Face Recognition Grand Challenge 3D facial database).

7)Local Binary Patterns Histograms (LBPH)

Local binary patterns (LBP), a straightforward, efficient texture operator in computer vision, are used in this technique to mark individual pixels in an image by setting a neighbourhood threshold for each pixel and then treating the result as a binary number. The LBPH method generates histograms for each labelled and classed image during the learning phase. Each image from the training set is represented by a different histogram. In this approach, comparing the histograms of any two photos is what the actual recognition procedure entails.

8)FaceNet

Based on benchmark datasets for face recognition, Google researchers created the FaceNet face recognition system in 2015. This system is quite well-known thanks to the readily available pre-trained models and multiple open-source third-party implementations. In comparison to other earlier-developed algorithms, FaceNet has good outcomes in research-conducting surveys, testing performance, and accuracy. FaceNet efficiently extracts face embeddings, superior features utilised later in the development process to train face recognition algorithms.

Summary

Numerous facial recognition algorithms and techniques exist. Although they all share a common primary goal, they can vary depending on the task or the issue. They range from neural networks and mathematical models to proprietary tech solutions, depending on the uses and implementation situations.

These most popular algorithms and techniques were addressed in this article. However, additional studies and scientific tests demonstrate the indisputable advantages of integrating several algorithms to improve facial recognition outcomes. It causes the emergence of novel approaches and processes tailored to certain uses.

Now there is a world's simplest facial recognition api for Python and the command line.The face_recognition command lets you recognize faces in a photograph or folder full for photographs. There's one line in the output for each face. The data is comma-separated with the filename and the name of the person found.

To know more about face_recognition module https://github.com/ageitgey/face_recognition