Facial Action Coding and Hybrid Deep Learning Architectures for Autism Detection

Abstract

Hereditary Autism Spectrum Disorder (ASD) is a neuron disorder that affects a person's ability for communication, interaction, and also behaviors. Diagnostics of autism are available throughout all stages of life, from infancy through adolescence and adulthood. Facial Emotions detection is considered to be the most parameter for the detection of Autismdisorders among the different categories of people. Propelled with a machine and deep learning algorithms, detection of autism disorder using facial emotions has reached a new dimension and has even been considered as the precautionary warning system for caregivers. Since Facial emotions are limited to only seven expressions, detection of ASD using facial emotions needs improvisation in terms of accurate detection and diagnosis. In this paper, we empirically relate the facial emotions to the ASD using the Facial Action Coding Systems (FACS) in which the different features are extracted by the FACS systems. For feature extraction, DEEPFACENET uses the FACS integrated Convolutional Neural Network (FACS-CNN) and hybrid Deep Learning of LSTM (Long Short-Term Memory) for the classification and detection of autism spectrum disorders (ASD). The experimentation is carried out using AFFECTNET databases and validated using Kaggle Autistic facial datasets (KAFD-2020). The Multi-Layer Perceptron (48.67%), Convolutional neural networks (67.75%), and Long ShortTerm Memory (71.56), the suggested model showed a considerable increase in recognition rate (92%), from this proposed model prove its superiority in detecting autistic facial emotions among children effectively.

Refurbished and improvised model using convolution network for autism disorder detection in facial images

“…Visual geometry group, also known as visual geometry group (VGG), is one of model of the convolution neural network. It is typical deep convolution network where "deep" indicate the number of layers with 16 or 19 convolution layers [7]. The architecture of VGG Architecture is shown in The input visuals are processed first, then these weighted layers are applied [11], [12].…”