Facial recognition algorithm.
This is an approach in technology field that deals with learning and recognition of face and identification of face from an image sent to the system. This is mostly applied in security measures (Naik, M.K. and Panda, R., 2016).
There are numerous method of identifying a face. The algorithm can use data to get a configuration which characterizes a particular individual.
Operation modes of face recognition systems.
- Verification/ authentication of facial images. It matches the test facial image with that saved in the system to grant permission.
- Identification. It compares the test facial image with the one stored in the system with the goal of discovering the user that matches that face.
Stated bellow are several common categories of face recognition algorithms.
Fisherface algorithm.
This is a type of algorithm that is used for face recognition. It is among the most common algorithms used for this purpose. This algorithm is believed to stand out among the other techniques such as eigenface, a superiority brought about by the fact that it employs more determination to exploit the difference between periods in the training (to recognize success of the system created) process. The process of face recognition using this algorithm is centered on reduction of space dimension by use of a technique known as Principal Component Analysis (PCA). Afterwards, Linear Discriminant Analysis (LDA) method is engaged to acquire feature of image characteristics. This type of face recognition is designed and developed to identify the face image by corresponding the outcomes of trait extraction (Gangopadhyay, I., Chatterjee, A. and Das, I., 2019. The system is then obligated to find out if the face image under test is identified appropriately or otherwise. During processing of images, the algorithm is employed to produce trait vector of facial image data and then compare vector of characteristics of the exercising image using vector features of an image under test using formula known as Euclidean distance. This algorithm is capable of identifying the face testing image of acceptably with unity which is one hundred percent.
Eigenface algorithm.
This is another method of face recognition algorithm that is being used with the sole purpose of recognizing each individual different from each other. In this algorithm, during the recognition process, an eigenface is made for the face image. Euclidian distances of this eigenface compared to that which is previously stored in the system are calculated.in analyzing the information from Euclidian distances, the eigenface with the least Euclidian distance is the one that has close resemblance to the person (Wagh, P., Thakare, R., Chaudhari, J. and Patil, S., 2015). The results are simulated mostly using Matlab program.
Local Binary Pattern Histogram (LBPH).
This algorithm is among the earliest face recognition known to man. Local binary pattern can be described as an effective quality operator that tags the pixels of an image by thresholding the locality of each pixel and reflects the output in form of a binary number. The images are characterized with simple data vector when associated with use of histograms. This algorithm appreciates the following constraints, radius for the middle pixel which is commonly set as one, neighbors normally set to eight, grid Y and grid X which are both set to eight. In conclusion to be able to identify the image which corresponds with the test image, a comparison is made between two histograms. An image that has the closest histogram is identified as a likely image. The comparison is based on the distances between the histograms that can be calculated by, chi-square and absolute value among many other methods (Acharya, U.R., Ng & W.L., Rahmat, K., Sudarshan., 2017).
All this methods are aimed at the same goal of face recognition. However, each algorithm has a different approach from each other. Below is a comparison between the above sated algorithms. The following are images of the apps that use LBPH algorithm.
Fig. 1.0: face recognition App
Fig. 1.1: Face recognition App ready for scanning
Fig 1.2: Face recognition App
Fig. 1.3: Face recognition App on phone
Fig 1.4: Face recognition App
Fig. 1.5 Face App installed
Fig. 1.6: fully installed on phones
Strong points.
| Eigenface algorithm | Fisherface algorithm | Local binary pattern histogram. |
| In this type of algorithm, no acquaintance of geometry as well as reflectance of faces is necessary. | As compared to eigenface algorithm, this type of algorithm has marked improvement in classification of images. | There is use of very high discriminative supremacy. |
| The principal of low dimensional subspace representation is employed to achieve appropriate data density. | This approach has an advantage of being extra invariant to light intensity. | There is simplified methods of computing for the histogram distances of different images. |
| Face recognition is simplified and very effective in relation to other types of algorithms. | There is assurance of improved accuracy in facial expression as compared to other approaches example eigenface approach. | This is considered one of the easiest face recognition algorithms to use because there is demonstration of local features in the image. |
| There is direct use of raw statistics for face recognition and learning without any substantial mid-level processing. | The approach is not affected byNoise included images as well as the blurring effect on the image. | There is assurance of decent performance in face recognition. |
| It is easy and quick to implement | The approach is easy and simple to learn. | The algorithm does not change with the variation of greyscale. |
| There are straightforward stages of face recognition. |
Weaknesses.
| Eigenface algorithm | Fisherface algorithm | Local binary pattern histogram. |
| This type of algorithm is considered to be extremely sensitive to scale therefore for the purposes of scale standardization, there is a necessity of mid-level processing. | This algorithm is more intricate than eigenface in locating the projection of face space. | The algorithm is subject to variation in rotation of images to be recognized. |
| The method is very sensitive to location and lighting of head. | The processes involved in fisherface outcomes to large storage space for the image under test and also more processing time is needed for this method. | Increase of the number of neighbors leads to increase of the size of the feature which results to an increase of computational density in relation to time and space. |
| Recognition rate is subject to variation of pose and illumination. The rate decreases under different conditions of both pose and illumination. | First time learning is very time consuming. | There is a constraint of operational data taken by the algorithm since only pixel difference is used, disregarding the size of information. |
| The method may need even background of images which may not be fulfilled in most natural scenes. | The approach does not easily appreciate changes in face arrangement and facial expression. | |
| This method can be categorized as a method that is very time consuming when it comes to learning it. This limits the update of face database. |
Below is an illustration of local binary pattern histogram face identification process.
References
Naik, M.K. and Panda, R., 2016. A novel adaptive cuckoo search algorithm for intrinsic discriminant analysis-based face recognition. Applied Soft Computing, 38, pp.661-675.
Gangopadhyay, I., Chatterjee, A. and Das, I., 2019. Face Detection and Expression Recognition Using Haar Cascade Classifier and Fisherface Algorithm. In Recent Trends in Signal and Image Processing (pp. 1-11). Springer, Singapore.
Wagh, P., Thakare, R., Chaudhari, J. and Patil, S., 2015, October. Attendance system based on face recognition using eigen face and PCA algorithms. In 2015 International Conference on Green Computing and Internet of Things (ICGCIoT) (pp. 303-308). IEEE.
Acharya, U.R., Ng, W.L., Rahmat, K., Sudarshan, V.K., Koh, J.E., Tan, J.H., Hagiwara, Y., Gertych, A., Fadzli, F., Yeong, C.H. and Ng, K.H., 2017. Shear wave elastography for characterization of breast lesions: Shearlet transform and local binary pattern histogram techniques. Computers in biology and medicine, 91, pp.13-20.
