Authors: Mr. Aditya Kulkarni, Mr. Parth Joshi, Mr. Shaunak Sindgi, Mr. Shreyas Rakshasbhuvankar, Mr. Vivek Kumar, Prof. Madhavi Dachawar
DOI Link: https://doi.org/10.22214/ijraset.2022.48268
Certificate: View Certificate
Age and gender predictions of unfiltered faces classify unconstrained real-world facial images into predefined ages and gender. Due to its value in intelligent real-world applications, this study topic has undergone significant advancements. Nevertheless, traditional methods based on unfiltered benchmarks have proven inefficient at handling large degrees of variation in unconstrained images. Because of their superior performance, Convolutional Neural Networks (CNNs) based approaches have been employed widely in recent years for the classification of jobs, and good quality of performance in facial analysis. In this work, we propose a novel end-to-end CNN approach, to achieve robust age group and gender classification of natural real-world faces. Two-level CNN architecture includes feature extraction and classification itself. The feature extraction process extracts a feature corresponding to age and gender, and the classification process classifies the face images according to age and gender. Particularly, we address the large variations in unfiltered real-world faces with a robust image pre-processing algorithm that prepares and processes those facial images before being given into the CNN model.
Facial Analysis has gained tremendous popularity in the field of computer innovation in recent years. A person's face contains characteristics that determine their age, gender, emotion, and ethnicity. It is considered that Gender and Age are top-tier classifications that are helpful for real-world applications in security and surveillance systems, ECMS (Electronic Customer Management Systems), Biometrics, Human-Computer Interaction, entertainment, and Forensic Art. There are still a few issues in gender and age grouping that remain unresolved. Despite the progress being made by the computer vision community with the continuous development and improvement of new techniques, age, and gender predictions of unfiltered real-life faces square measure nevertheless to satisfy the wants of commercial and real-world applications. Over the few years, a lot of routes suggested solving the grouping problem. Most of those routes are handcrafted which results in undesirable performance on the age and gender predictions of unconstrained in-the-wild image. These traditional hand-engineered routes relied on the distinct dimensions of facial features and face signifiers which cannot handle diverse degrees of disparity perceived in these exigent ungoverned imaging stipulations. Images in this category might differ due to a few disparities in appearance, noise, pose,
And lighting may affect the manually designed computer vision routes to at par grouping of age and gender images. Over the past few years, a variety of approaches are instructed to resolve the grouping drawback. Most of those routes are manually constructed, which results in subpar performance on the age and gender predictions of unconstrained in-the-wild images. Tradition hand-engineered routes were relying on facial characteristics and signifiers that do not have the capacity to deal with differences in degrees of disparity perceived in these ungoverned, exigent imaging conditions. Images in this category may differ due to minor differences in appearance, noise, pose, and lighting that affect the route manually designed for grouping images by age and gender.
II. RELATED WORK
In this section, we briefly review the age and gender classification literature and describe both the early methods and those that are most related to our proposed method, focusing on age and gender classification of face images from unconstrained real-world environments. Almost all the early methods for age and gender classification were handcrafted. Based on constrained images that were taken under controlled imaging conditions, they manually engineered facial features from faces.
To name a few, Kwon and Lobo  developed an initial method for estimating age based on geometric features of the face in 1999. The ratios between the different dimensions of facial features are determined by these features. Geometric features successfully distinguish babies from adults, but are unable to distinguish between young adults and senior adults. Therefore, Lanitis et al.  proposed an Active Appearance Model (AAM) based estimation method that incorporates both spatial and texture features. As a result, it is not suitable for the unconstrained imaging conditions associated with real-world face images, which are subject to variations in illumination, expression, poses, etc. From 2007, most of the approaches also employed manually designed features for the estimation task: Gabor , Spatially Flexible Patches (SFP), Local Binary Patterns (LBP), and Biologically Inspired Features (BIF). In recent years, classification and regression methods have been employed to classify the age and gender of facial images according to those features. Classification methods in use Support Vector Machine (SVM) based methods for age and gender classification. Several regression methods can be implemented to predict age and gender, including linear regression, support vector regression (SVR), Canonical Correlation Analysis (CCA), and partial least squares (PLS). Dileep and Danti also developed a method that utilized feedforward neural networks and a 3-sigma control limits the approach to classify people’s age into children, middle-aged adults, and old-aged adults. However, all these methods are only suitable and effective on constrained imaging conditions; they cannot handle the unconstrained nature of real-world images and, therefore, cannot be relied on to achieve respectable performance on the images which are common in practical applications.
III. PROPOSED METHOD
This section presents the proposed deeply learned classifiers for age group and gender classification of unfiltered real-life face images, an extension of our conference paper . The approach as presented in Algorithm 1 requires an image pre-processing (face detection, landmark detection, and face alignment) stage that pre-processes and prepares the face images before they are input into the proposed network. Therefore, our solution is divided into three major steps: image pre-processing, feature learning, and classification itself.
2. Face Detection: The first stage of image pre-processing is face detection. The face detection state locates the face in an input image. In this work, we employ an open-source face detector: Head Hunter described in . In order to detect a face, all the input pictures square measure revolved within the range of −90° to 90° angles and with a step of 5°. After that, the detector confirms the input image with the most accurate output of the face detector and in an exceedingly case wherever the face is not the modifications of the input image, the original input image is heightened and the face detection algorithm is replicated until a face is conceded. An upscaling technique helps in discovering faces in all the input images.
3. Image Landmark & Face Alignment: A subset of face detection is the facial landmark detection and face alignment stage, where we utilize the state-of-the-art solution in , is image pre-processing solution is an open-source Multiview facial landmark detection algorithm that uses five landmark detection models, including a frontal model, two half-profile models, and two full profile models. All these five models are trained to work in one of the correlated portrait poses. Face alignment, on the other hand, requires running all five facial landmark models on the detected faces. An affine transformation is then performed on the model, with the highest confidence score, to the predefined optimal settings of those landmarks.
4. CNN Architecture: Our CNN architecture is a novel six-layer network, comprising four convolutional and two fully connected layers. The CNN design is an end-to-end sequential deep learning architecture, including feature extraction and classification phases. The feature extraction phase has four convolutional layers, with the corresponding parameters, including the number of each filter's kernel size, as well as the number of filters, and the stride. It consists of a convolutional layer, activation layer, batch normalization, max-pooling layer, and drop out. On another hand, the classification layer contains two well-connected layers, that deal with the classification stage of the model. We arranged the Gender and Age grouping task as an end-to-end deep classification problem; hence, SoftMax with a cross-entropy function is embraced to acquire a probability for each age group and gender class. We show the layout of some parameters of CNN architecture in the section below. SoftMax gives you the probability for each class label below;
5. Training Details: In this very section, we are going to describe the training details for age group and gender classifiers on our datasets benchmark. The age group classifier shall be at authority for predicting the age groups of natural or raw person’s face images into eight different classes, while the gender classifier shall differentiate those facial images into two distinct gender classes. For all our experiments, we initialize and trained our CNN model from scratch using the images and the labels of the datasets benchmark. We primarily pre-trained the novel CNN architecture on the natural facial aging dataset whose images are acquired from the web with a distinct level of capriciousness and then calibrate the CNN on the images from the dataset, to prevent overfitting and to refashion the CNN model of face images to perform well.
IV. PROPOSED ALGORITHM
The proposed algorithm for the CNN model is given below in Algorithm 1.
Face Detection: The above diagram shows the flow of the working of the novel CNN model and how it would be predicting the gender and classify it accordingly to the respective sex. Inventive writing, facial uncovering, and acknowledgment are congruously most of the vastly scrutinized themes. The rudimentary stimulus behind face recognition and acknowledgment of the facial personalized picture which consists of distinct testing issues. To gather the Framework there are an exceptional deal of difficulties thinking about the varieties in present, lights, impediments and revolution of the photo, scaling element, and appearance of the face.
In this element, the exploratory aftereffects of the utilization are brought. we've as of now tested this system works inside the machine segments. In this segment, we have got given the subtleties about the effects obtained via us, while making use of this program against the absolute experiments. we have clarified the yield of every single experiment making use of the display screen captures of the yield given by way of our program. even as making this task, we faced a top-notch deal of difficulties and we have got tried to restrict it however a lot as can be expected of those problems. We assessed the proposed calculation on the steady video outlines.
VI. FUTURE SCOPES
An AI software program application that is used to locate the age and gender of users who passes through based on line face analyses and automatically starts off evolved playing commercials based totally on the centered audience. An Android app that determines your age from your photographs using facial popularity. it may bet your age and gender alongside that also can locate multiple faces in a photo and estimate the age of each face.
To understand an individual, during this paper we have got proposed each projected Convolution Neural Network (CNN) and near Binary patterns Histograms (LBPH) technique for extracting the functions and matching the manner for face detection, pursuit, and popularity. observe that the identical technique has been applied to the other challenge that builds on localization, consisting of face watching, and face detection severally and the identical CNN methodology is applied to gender estimation severally. We have got 1st proven that trendy measures employed in face detection, pursuit, and recognition mutually.
 Agbo-Ajala O, Viriri S. Deeply Learned Classifiers for Age and Gender Predictions of Unfiltered Faces. ScientificWorldJournal. 2020 Apr 30; 2020:1289408. doi: 10.1155/2020/1289408. PMID: 32395084; PMCID: PMC7201854.  E. Agustsson, R. Timofte, S. Escalera, X. Baro, I. Guyon, and R. Rothe, “Apparent and real age estimation in still images with deep residual regressors on the appa-real database,” in Proceedings of the 2017 12th IEEE International Conference on Automatic Face & Gesture Recognition (FG 2017), Biometrics Wild, Bwild, Washington, DC, USA, pp. 87–94, June 2017.  K. Zhang, C. Gao, L. Guo et al., “Age group and gender estimation in the wild with deep RoR architecture,” IEEE Access, vol. 5, pp. 22492– 22503, 2017.  A. Kuehlkamp, “Age estimation from face images,” in Proceedings of the 6th IAPR International Conference on Biometrics (ICB), pp. 1–10, Madrid, Spain, June 2013.  V. Carletti, A. S. Greco, G. Percannella, M. Vento, and I. Fellow, “Age from faces in the deep learning revolution,” IEEE Transactions on Pattern Analysis and Machine Intelligence, p. 1, 2019.  B. Bin Gao, H. Y. Zhou, J. Wu, and X. Geng, “Age estimation using expectation of label distribution learning,” in Proceedings of the Twenty-Seventh International Joint Conference on Artificial Intelligence, pp. 712–718, Stockholm, Sweden, July 2018.  R. C. Malli, M. Aygun, and H. K. Ekenel, “Apparent age estimation using ensemble of deep learning models,” in Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition Workshops, pp. 714– 721, Las Vegas, NV, USA, June 2016.  G. Antipov, M. Baccouche, S. A. Berrani, and J. L. Dugelay, “Effective training of convolutional neural networks for facebased gender and age prediction,” Pattern Recognition, vol. 72, pp. 15–26, 2017.  J. G¨unther, P. M. Pilarski, G. Helfrich, H. Shen, and K. Diepold, “First steps towards an intelligent laser welding Scientific World Journal architecture using deep neural networks and reinforcement learning,” Procedia Technology, vol. 15, pp. 474–483, 2014.  C. Szegedy, W. Liu, Y. Jia et al., “Going deeper with convolutions,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 1–9, Boston, MA, USA, June 2015.  J. Huang, B. Li, J. Zhu, and J. Chen, “Age classification with deep learning face representation,” Multimedia Tools and Applications, vol. 76, no. 19, pp. 20231–20247, 2017.  E. Eidinger, R. Enbar, and T. Hassner, “Age and gender estimation of unfiltered faces,” IEEE Transactions on Information Forensics and Security, vol. 9, no. 12, pp. 2170–2179, 2014.  G. Levi and T. Hassncer, “Age and gender classification using convolutional neural networks,” in Proceedings of the 2015 IEEE Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), pp. 34–42, Boston, MA, USA, June 2015.  F. Gao and H. Ai, “Face age classification on consumer images with gabor feature and fuzzy LDA method,” in Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 5558, pp. 132–141, LNCS, Springer Science+Business Media, Berlin, Germany, 2009.  A. G¨unay and V. Vasif, “Age estimation based on AAM and 2D-DCT features of facial images,” International Journal of Advanced Computer Science and Applications, vol. 6, no. 2, pp. 113–119, 2015.  A. G¨unay and V. Vasif, “Facial age estimation based on decision level fusion of AAM, LBP and gabor features,” International Journal of Advanced Computer Science and Applications, vol. 6, no. 8, pp. 19–26, 2015.  S. Yan, M. Liu, and T. S. Huang, “Extracting age information from local spatially flexible patches,” in Proceedings of the IEEE International Conference on Acoustics, Speech and Signal Processing, pp. 737–740, Las Vegas, NV, USA, March 2008.  M. Y. El Dib and M. El-saban, Human Age Estimation Using Enhanced Bio- Inspired Features (EBIF), Faculty of Computers and Information, Cairo University, Cairo, Egypt, 2010.  Y. H. Kwon and N. Da Vitoria Lobo, “Age classification from facial images,” Computer Vision and Image Understanding, vol. 74, no. 1, pp. 1– 21, 1999.  A. Lanitis, C. Draganova, and C. Christodoulou, “Comparing different classifiers for automatic age estimation,” IEEE Transactions on Systems, Man and Cybernetics, Part B (Cybernetics), vol. 34, no. 1, pp. 621– 628, 2004.
Copyright © 2022 Mr. Aditya Kulkarni, Mr. Parth Joshi, Mr. Shaunak Sindgi, Mr. Shreyas Rakshasbhuvankar, Mr. Vivek Kumar, Prof. Madhavi Dachawar. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
Authors : Parth Joshi
Paper Id : IJRASET48268
Publish Date : 2022-12-20
ISSN : 2321-9653
Publisher Name : IJRASET
DOI Link : Click Here