When performing facial recognition for an individual in your dataset, it is necessary to know the workflow process of facial recognition. Security systems are essential for everyday life, to control access rights only to certain people. Traditional security systems include keys, passwords, or ID cards. The weakness of these systems is that they are difficult to remember, can be lost, and can even be easily known by others.
What is Face Recognition
Based on these weaknesses, facial recognition technology has emerged. Facial recognition is known as one of the image processing technologies that can be applied in the field of security systems. So, what is facial recognition? Facial recognition is one of the biometric technologies used to identify or recognize an individual based on certain characteristics. Facial recognition is considered to have a high level of security because facial images are difficult to imitate, modify or steal compared to traditional security systems.
How Does Facial Recognition Workflow Process Work?
Have you ever imagined how a system can recognize an individual with high accuracy? With the many machine learning and artificial intelligence technologies, and with certain methods, this is very possible. Here is the Facial Recognition Workflow Process:
In its implementation, there are several stages in the facial recognition process, including database training and individual matching:
Training Database
The process of training facial recognition (facial identification) images usually involves several steps, including:
Image Acquisition
Face Image Acquisition is the process of capturing or scanning an analog image to obtain a digital image. Some factors that need to be considered in the image acquisition process include the type of acquisition device, camera resolution, lighting techniques, zooming, distance, and angle of image capture.
Facial Image Data Collection
The next step in the facial recognition training process is to collect a facial image dataset that will be used to train the model. This dataset must consist of facial images represented by feature vectors or relevant facial features that can differentiate one face from another.
Pre-Processing
After the data is collected, facial images are usually processed to remove noise or irrelevant information. Pre-processing is one of the stages in facial recognition where facial image data will go through the process of cropping, face detection, resizing, and changing RGB format to grayscale. The aim of the pre-processing stage is to enable better processing of facial images and increase the system’s likelihood of successfully identifying faces quickly.
Cropping
Cropping is a technique used to determine which part of a facial image contains the object area that will be processed in the next stage, so that it can be cut and separated from the unwanted area.
Face Detection
Face Detection is a stage to detect only the facial part of an individual so that a simpler image output is produced. Face Detection in Facial Recognition generally uses various methods. One of the most commonly used face detection methods is the Viola-Jones method. This method aims to remove parts that are not identified as faces. The goal of face detection is to improve system performance in matching facial images.
Resize
Resize is the process of changing the pixel size of a facial image. Each image resulting from face detection has different pixel sizes, so it is necessary to standardize the image size to facilitate the system in recognizing individuals and generating a database of images with the same size.
RGB to Greyscale
Next, the RGB image will be converted to a grayscale image. The function of this stage is that grayscale images have a simpler structure and facilitate the computation process. The hope is that the facial matching process will be faster and more efficient than RGB images.
Feature Extraction
Feature extraction is a process of capturing the characteristics of an object that distinguishes it from other facial images in a dataset. It is achieved through the use of face detection algorithms such as Local Binary Patterns (LBP), Principle Component Analysis (PCA), Eigenface, Histogram of Oriented Gradients (HOG), or Convolutional Neural Networks (CNN).
In this stage, a facial image is represented as a matrix in the system. The matrix values are then averaged to compare one image with another. Here is an example of how feature extraction is performed using the Eigenfaces algorithm, which is a result of the PCA process used in face recognition. Eigenfaces consist of feature vectors that represent the common characteristics or patterns of faces in a facial image dataset. The following are the steps involved in feature extraction using the eigenface method.
Model Formation
After facial features are extracted from the facial image dataset, a face recognition model can be formed using machine learning algorithms such as Support Vector Machine (SVM), Random Forest, or Convolutional Neural Network (CNN). In this stage, machine learning algorithms are applied to the training data and a model that is appropriate for the data is found. This model is generated using specific machine learning techniques to find correlations and patterns in the data that were taken from the preprocessing and feature extraction stages.
Training and Model Evaluation
The final step is to train the model using the preprocessed facial image dataset and the model formation stage. The model is then evaluated using a different testing dataset to ensure optimal performance and good generalization to new facial data.
Individual Recognition or Matching
The process of facial recognition (facial identification) usually involves several steps, including:
Matching Process
In this stage, a new facial image or test image is attempted to be recognized. After the facial features have been successfully extracted, the next step is to compare these features with the facial features that are already stored in the database. The common method used is the Euclidean distance method or cosine similarity.
Euclidean Distance is a method of measuring the straight line distance between two points, for example, point X (X1, X2, …, Xn) and point Y (Y1, Y2, …, Yn). The Euclidean Distance function is used to group data by calculating the distance between data points. The smaller the resulting distance value, the more similar the data can be classified into one group (having similarity with facial images in the dataset).
The Euclidean value is used to calculate the distance for each training data one by one, to find the threshold value. The threshold value is the minimum distance that must be passed by the testing data, to prove that the testing data is recognized or is present in the database. If the testing facial image has a value greater than the minimum distance set, then the testing facial image is not recognized or not present in the database.
Decision
In the final stage, the system will make a decision based on the matching results from the previous stage, for example, “recognized” or “not recognized.” From this decision, you can connect it to other output results such as sending binary data to a microcontroller component for locking and unlocking systems and others.
Conclusion
In face recognition, many things affect the accuracy of the system, such as light intensity, image acquisition distance, and the angle of the facial image, which also greatly affects the matching results of an individual. This can be minimized by using more relevant algorithms and technology currently available. The preprocessing stage of the image is also very important in determining the success of the face recognition system. Therefore, each stage in the process must be carried out carefully and accurately to produce accurate and satisfactory results.
Writer : Meilina Eka A
