CNN-based Gender Prediction in Uncontrolled Environments

Yildiz, Kazım; Güneş, Engin; Baş, Anıl

doi:10.29130/dubited.763427

Cited by 7 publications

(3 citation statements)

References 31 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The maximum accuracy was 90% when using the Nottingham Scan database and 97.44% when using the Kaggle dataset. In a study by Yildiz et al (2021), CNN's algorithm was trained using facial images that varied greatly regarding race, position, age, and illumination. The authors created low-resolution images using bilinear interpolation to create an approximate real-world photography scenario.…”

Section: Paper Contributionsmentioning

confidence: 99%

“…Comparing the proposed method with some of the previous techniques for gender classification is listed in Table 12 for face images and Table 13 for eyes images. (Sumi et al, 2021) Kaggle dataset, Nottingham scan 97.44,90 Deep CNN (Yildiz et al, 2021) Adience, VGGFace2 85.52,93.71 Pareto frontier CNN (Islam et al, 2020) WIKI-cleaned 90 Pre-trained CNN (Zhou et al, 2019) Adience 93.22 VGGNet arch (Dhomne et al, 2018) Celebrity faces 95 Hyper face (Ranjan et al, 2017) CelebA, LFWA 98,94 Face tracer (Kumar et al, 2008) CelebA, LFWA 84,91 Deep CNN (Kamaru, 2020) CelebA, LFWA 96,95 Deep CNN (Benkaddour et al, 2021) WIKI, IMDB 93.56,94.49 CNN + ELM (Extreme Learning Machine) (Micheala and Shankar, 2021) Adience 90.2 LMTCNN (Lightweight Multi-task CNN) (Lee et al, 2018) Adience 85 Wide CNN + Gabor Filter (Hosseini et al, 2018) Adience 88.9 2DPCA on real Gabor space + SVM (Rai and Khanna, 2015) LFW…”

Section: Train the Cnn Model On Eyes From Different Datasetsmentioning

confidence: 99%

See 1 more Smart Citation

Gender Classification from Face and Eyes Images Using Deep Learning Algorithm

Ghrban¹,

Abbadi²

2023

Journal of Computer Science

View full text Add to dashboard Cite

Gender classification provides additional information about the individual's identity, which is crucial in surveillance, smart interface, and smart advertising, it is a very important aspect of face analysis that has piqued the interest of researchers in areas such as demographic information collection, surveillance, human-computer interaction, marketing intelligence, security, etc. Usually, facial images are used to extract features for classification. This study aims to determine gender based on the entire face or the eyes (for masked or occlusion faces). The proposed method consists of six main stages: (1) Remove the background using the Deep Labelling Version 3 plus (DeepLabV3+) method; (2) Skin detection by applying the combination of the two colors spaces models (Hue, Saturation, Value (HSV)) and (Luminance, Chrominance blue and Chrominance red (YCbCr)); (3) Face detection using the Haar Cascades classifier method; (4) Face alignment and cropping; (5) Classify the gender. Classification of the gender when the image contains the entire face is based on a deep wavelet. In the case of occlusion faces, we proposed to use a Convolution Neural Network (CNN) for classifying the gender from the eye (s). The dataset used for training this model is the celeb faces attributes dataset (CelebA) and some of the different datasets were also used to compare this model with the other previous works. In addition, we showed that our models perform well with images with some challenges, for example, some of the faces are not fully visible, not completely frontal, wearing glasses with different styles, have low quality and noise, have various lighting, children's and infants' faces, with and without makeup, open and closed mouths and closed eyes. Other than that, the accuracy achieved was 98% when classifying from face images and 98% from the eye. The proposed method was compared with previous works and was very promising. The contributions of this proposal were the use of the deep wavelet for gender classification and the proposal of a new method for skin detection, which enhances the performance of the Haar cascade method used for face detection. Also, we proposed a method for face alignment and, finally, the classification of genders with many challenges from the entire face or just from the eye (s).

show abstract

Section: Paper Contributionsmentioning

confidence: 99%

Section: Train the Cnn Model On Eyes From Different Datasetsmentioning

confidence: 99%

Gender Classification from Face and Eyes Images Using Deep Learning Algorithm

Ghrban¹,

Abbadi²

2023

Journal of Computer Science

View full text Add to dashboard Cite

show abstract

“…An automated FER system can be seen as a supervised classification method comparing selected facial features from given image or video frame with faces within a database. It is a well established fact that computer vision tasks are optimally solved by convolutional neural network (CNN) and, it is usually necessary to have large databases in order to avoid overfitting [11][12] [13]. Unfortunately, some public image-labeled databases used to train and test FER systems, such as Karolinska Directed Emotional Faces (KDEF) [14] and Extended Chon-Kanade (CK+) [15], are not sufficiently large.…”

Section: Introductionmentioning

confidence: 99%

Data Augmentation and Transfer Learning Approaches Applied to Facial Expressions Recognition

Randellini¹,

Rigutini²,

Saccà³

2021

NLP Techniques and Applications

View full text Add to dashboard Cite

The face expression is the first thing we pay attention to when we want to understand a person’s state of mind. Thus, the ability to recognize facial expressions in an automatic way is a very interesting research field. In this paper, because the small size of available training datasets, we propose a novel data augmentation technique that improves the performances in the recognition task. We apply geometrical transformations and build from scratch GAN models able to generate new synthetic images for each emotion type. Thus, on the augmented datasets we fine tune pretrained convolutional neural networks with different architectures. To measure the generalization ability of the models, we apply extra-database protocol approach, namely we train models on the augmented versions of training dataset and test them on two different databases. The combination of these techniques allows to reach average accuracy values of the order of 85% for the InceptionResNetV2 model.

show abstract

An Evaluation of the Performance of Convolution Neural Network and Transfer Learning on Face Gender Recognition

Babatunde,

Ogundokun

et al. 2023

Lecture Notes in Networks and Systems

View full text Add to dashboard Cite

CNN-based Gender Prediction in Uncontrolled Environments

Cited by 7 publications

References 31 publications

Gender Classification from Face and Eyes Images Using Deep Learning Algorithm

Gender Classification from Face and Eyes Images Using Deep Learning Algorithm

Data Augmentation and Transfer Learning Approaches Applied to Facial Expressions Recognition

An Evaluation of the Performance of Convolution Neural Network and Transfer Learning on Face Gender Recognition

Contact Info

Product

Resources

About