Enhanced Face Recognition System: Integrating of Collaborative Representation Based Classification (CRC) _KNN

1. Introduction

A few decades ago, the concept of face recognitionwasgaining a lot of attention due to the major contributions towards the emerged security systems and mobile robotics. The technique of face recognition, enhanced the capability of various system specifically to perform difficult tasks.Earlier, these difficult tasks are performed by the extreme effort of humans.The representative cases are the security systems of various apartments, airports and agencies. These systems are obligatory to be performed in a secure manner and have the reliability to activate in real-time systems. A basic conventional face recognition system achieved high accuracyrate, but the systems are expensive is explaned by Arya and Agrawal (2018). Such as the face recognition method of deep learning that motivated the researchers to get a better solution for the face recognition which is quick as compared to other methods.Thus, a new planned algorithm is applied for the real-time face detection using face recognition system.In the face recognition system a major issue needs to be resolved, which is occurring because of the restriction over the number of training phases. The gathering of large numbers of training faces is not considered as a simple task. So, SRC (Sparse representation based classification) is required to fulfill the need. Nowadays it is gaining attention due to the description of facial feature vectors on the overall datasets. The progression is effective, but it has a high cost that makes it unreliable to various security systems as described by Wright, Yang, Ganesh, Sastry, and Ma (2009). Presently, CRC (Collaborative representation based classification) becomes a trendy algorithm for the face recognition system. The obtained solutions given by Rahim, Afriliansyah, Winata, Nofriansyah, and Aryza (2018) are better as compared with the performance of SRC in terms of accuracy and the procedure of classification is performed in a short time period.

As stated by Zhang, Yang, and Feng (2011) the algorithm has certain shortcomings related to the training faces and these are minimized by the use of another algorithm such as KNN (K nearest neighbors).To sort out all the major issues the primitive technique isa Collaboration classification CRC-KNN. CRC is a fascinating concept for researchers. The CRC classification is effective as well as giving high accuracy rates as compared with the conventional representation based classification in the process of face recognition.The accuracy is degraded if the training faces aresmall in each class. It happened because CRC reliant on the Euclidean distance. The performance is enhanced by grouping CRC with KNN. K nearest neighbor classification is globally recommended for the applications of machinelearning, feature extraction and also for the data mining purposes. The execution process and the performance of KNN is simple. The classification technique considered the K values by selecting a fix constant for the entire data and introduced cross-validation to determine the k values for different data points.The planned algorithms are determining the accuracy offace recognition systems in different environments. The contributions of new algorithms are described below:

• 1.

  The new planned algorithm overweigh the performance of an existing algorithm;

• 2.

  The purpose of the new algorithm is to recognize the faces which, depending upon the classification on the basis of both CRC and KNN;

• 3.

  A maxout network is trained to extract the features of the face for the recognition process. Next to it, a new method is accessed as namely as Collaboration classification CRC-KNN. It does not only sort out the major issues of existing algorithmswith high accuracy, but also make the system reliable for real-time applications;

• 4.

  The combined approachwon’t just fix the challenges, but also make it capable to perform better in different environments as discussed by Vo and Lee (2018) and Zhang et al. (2017).