Event Title

Age Detection Using Convolutional Neural Networks

Presenter Information

Chivin So, Lander UniversityFollow

Location

Breakout Session A: Computer and Data Sciences

CASB 102

Start Date

8-4-2022 1:45 PM

End Date

8-4-2022 2:00 PM

Description

Facial analysis from images has attracted a lot of attentiveness due to the fact that it helps solve many problems including targeting customers for a particular product, targeting ads for a specific group of customers, give more superior content recommendation as well as security field [1]. More importantly with age detection combined with facial recognition, it allows to track human activity and also monitor people’s health based on their age [3]. The main objective of this research is to detect their age from a facial image. Our research aims to develop an age classifier using Convolutional Neural Networks (CNNs) on TensorFlow. We are developing classification models using Jupiter notebooks on machine learning server provided by Lander University. CNNs are well-known to work tremendously well with classification tasks with image data [4]. Due to the nature of our classifier using facial images, we have chosen to employ CNNs in our models. For this project, the research team built and combined the dataset from various sources online. Some of the famous ones include IMDB-WIKI dataset, Labeled Faces in the Wild (LFW) dataset and All-Age-Faces dataset [2]. We combined the UTKFace images from github [5] with facial-age dataset from Kaggle [6] and a subset of IMDB-WIKI consisting of 500k+ face images with age and gender labels [7]. We have integrated in a total of 43623 images for our research project. In order to achieve higher accuracy, we make the dataset even larger by using image data augmentation technique that creates extra images by zooming in/out, rotating, shifting, as well as flipping the original images. Our CNN model is an artificial neural network that consists of an input layer that takes image data, multiples of convolutional, pooling, dropout layers for image abstraction, a flatten layer and a dense layer for an output. The model is trained with the training image set and provides a reasonable accuracy. The idea is to get an initial benchmark on the model’s performance, and then incrementally try different CNN model structures, hyperparameter settings and techniques and to improve the performance [2]. Alongside, we are also developing an app for demo that captures an image from a webcam to classify the user’s age range. References to be provided.

This document is currently not available here.

Share

COinS
 
Apr 8th, 1:45 PM Apr 8th, 2:00 PM

Age Detection Using Convolutional Neural Networks

Breakout Session A: Computer and Data Sciences

CASB 102

Facial analysis from images has attracted a lot of attentiveness due to the fact that it helps solve many problems including targeting customers for a particular product, targeting ads for a specific group of customers, give more superior content recommendation as well as security field [1]. More importantly with age detection combined with facial recognition, it allows to track human activity and also monitor people’s health based on their age [3]. The main objective of this research is to detect their age from a facial image. Our research aims to develop an age classifier using Convolutional Neural Networks (CNNs) on TensorFlow. We are developing classification models using Jupiter notebooks on machine learning server provided by Lander University. CNNs are well-known to work tremendously well with classification tasks with image data [4]. Due to the nature of our classifier using facial images, we have chosen to employ CNNs in our models. For this project, the research team built and combined the dataset from various sources online. Some of the famous ones include IMDB-WIKI dataset, Labeled Faces in the Wild (LFW) dataset and All-Age-Faces dataset [2]. We combined the UTKFace images from github [5] with facial-age dataset from Kaggle [6] and a subset of IMDB-WIKI consisting of 500k+ face images with age and gender labels [7]. We have integrated in a total of 43623 images for our research project. In order to achieve higher accuracy, we make the dataset even larger by using image data augmentation technique that creates extra images by zooming in/out, rotating, shifting, as well as flipping the original images. Our CNN model is an artificial neural network that consists of an input layer that takes image data, multiples of convolutional, pooling, dropout layers for image abstraction, a flatten layer and a dense layer for an output. The model is trained with the training image set and provides a reasonable accuracy. The idea is to get an initial benchmark on the model’s performance, and then incrementally try different CNN model structures, hyperparameter settings and techniques and to improve the performance [2]. Alongside, we are also developing an app for demo that captures an image from a webcam to classify the user’s age range. References to be provided.