Facial Sentiment Analysis using Deep Neural Networks

نوع: Type: thesis

مقطع: Segment: masters

عنوان: Title: Facial Sentiment Analysis using Deep Neural Networks

ارائه دهنده: Provider: Elham Afshar

اساتید راهنما: Supervisors: DR. Hassan Khotanlu

اساتید مشاور: Advisory Professors:

اساتید ممتحن یا داور: Examining professors or referees: DR. Muharram Mansoorizade & DR. Abbas Ramezani

زمان و تاریخ ارائه: Time and date of presentation: 22/10/2022

مکان ارائه: Place of presentation: seminar2 omran

چکیده: Abstract: Facial sentiment plays an important role in conveying concepts in human communication, so that researches have shown that 55% of concepts are conveyed through facial emotion and only 7% of it is conveyed by words and sentences. This topic made many researchers interested in the field of facial emotion analysis and recognition because this field can be used in many fields of machine vision, including human-computer interaction, emotional computing, etc. In recent years, due to the increasing progress of neural networks, many researches have been conducted in the field of facial sentiment analysis and recognition. In this research, a method based on ensumble classification using convolutional neural networks to analyze and recognize facial sentiment is presented. In the first neural network, the addition of the spatial features of the image to its general features has been used to create a feature map as the input of the classification stage. In the second network, the local binary pattern of the images is used as the input of the first network. Since the local binary pattern can extract the texture of the images well, it can be effective in recognizing the bumps and facial expressions in different facial sentiment. After training the two proposed networks, the maximum probability between the two networks is considered as the final output in order to classify the emotion. The proposed method has been applied and tested on the FER2013 dataset. The results obtained from testing the proposed method on the data set show that the model has performed well and compared to the previous methods, it has achieved acceptable results

فایل: ّFile: Download فایل