Article Preview
TopIntroduction
In a vision structure, human images are more static over the person’s life time. Hence identifying or recognizing images are not surprising capability for human being. At the same time, Yang et al (2017) states that it needs a glance of person’s face to think back about the particular person. Human brain is having particular part exclusively to identify and remember the images by recalling the key details of the face like nose, eyes, ears, mouth and cheeks. Ramkumar et al (2018) states that, the human brain can recognize the faces to some extent even the image has been affected by uncontrolled situations such as illumination, blur and pose variation. On the opposite side Yan et al (2017) mentioned that face recognition by computer is depends upon the predefined factors. If the face is affected by the fore said issues, then the data needed for a computer to recognize the face is increasing and the computing operations are also becoming tricky. In this regard, Yuan et al (2017) discuss that many machine learning algorithms were developed that includes Deep Neural Network, CNN, Deep Belief Network (DBN) and Artificial Intelligence (AI). Since, enormous development in face recognition technology gradually increasing the purpose of interacting with humans and recognition of face finds major part in many applications, Coskun et al (2017). Even though, there are many methods available for serving this purpose, we are in need of achieving more effective recognition with good accuracy and less computation time. That’s the reason, How the face recognition technology become a wide area of research that grabs the creative ideas of the researchers, Itqan et al (2016).
Tamil et al (2019) states that traditional algorithms which are existing for face recognition are not handling the images effectively and also they could not process the large sample size of data which is a highly needed requirement in the current scenario with the known thing that the entire globe is moving gradually towards the automation to implicate the various purposes of recognizing faces to serve wide variety of applications including bio metric identification, serving security applications, monitoring and tracking live stream video etc.
Topical developments in the field of automatic face recognition, design acknowledgment and AI have made it conceivable to create an effective face recognition method to overcome the issues faced in various aspects of face recognition, Wang et al (2017). From one viewpoint, perceiving face is common procedure, since most of the individuals can do it easily with less cognizant. Then again, doing the same recognition with the help of machine is a troublesome issue. There are many traditional algorithms are available for face recognition. Generally, face recognition is done by putting the face recognition algorithms on two broad categories. They are holistic approach and local feature approach, Tian et al (2018).
The appearance-based methods are generally linear in nature and coming under the category of holistic approach that includes principal component analysis (PCA), independent component analysis (ICA), linear discriminate analysis (LDA) etc, Rajesh et al (2020), Shen et al (2018). These methods are not sufficient to face the images that are affected by poor lighting since the most important key features of the images are available over the non-linear plane. Highlighting these issues for an effective face recognition, the main objective of this work is arrived that implies a hybrid frame work of EVB_CNN for an effective face recognition over uncontrolled conditions with enhanced accuracy and less computation time. In the proposed method, the face recognition is performed in two important steps. (i)Detecting and normalizing the face using the calculated Eigen values (ii)Face recognition by CNN where the feature extraction and classification done as a combined work. Finally, afore mentioned steps effectively recognizing the face and showing enhancement in accuracy rate.
The remaining part of this paper is structured as like this II Theoretical Framework, III Literature survey, IV Methodology, V Sample Selection, Related work, VI Findings and Discussion VII Conclusion and Recommendation.