Truth From Facial Expressions With Deep Learning Program That Detects Timely Emotion Analysis And Design
Meryem Özlem AYDOĞAN
~From the report content:
1.2 Similar Examples in the World and in Turkey
1.2 Similar Examples in the World and in Turkey
Real-time emotion detection has attracted a lot of attention in the field of artificial intelligence and image processing in recent years. For this reason, there are many similar examples in the world and in our country. When similar projects that can be accessed as a result of the research are examined, it is found that models that perform face recognition, gender classification and emotion classification simultaneously in a single step using real-time convolutional neural network architecture have created awareness. Applications that predict age based on facial expressions have been developed and large data sets have been used for these applications. It is known that platforms such as "Affdex" developed by MIT, Microsoft's "Azure Cognitive Services" and "Face API, Amazon "Rekognition", "SkyBiometry", Emotion AI", "Face++", "Anthropic" are working on real-time emotion detection projects. These projects use deep learning techniques to classify emotional responses by building sentiment analysis models using large datasets. Pang and Lee pioneered the machine learning method, which has the most effective results in the studies conducted in the field of emotion analysis in the literature. The best success rate was 78.9%.
In Pang and Lee's project, text classification was treated differently. Dandil and
Özdemir, 2019, in their study, the classical convolutional neural network AlexNet and real
proposed an emotion recognition system based on facial expressions from timed video frames.
Chen and Ark, in their 2015 study, proposed an emotion classification system based on
with the automatic learning convolutional neural network (ESA-CNN) that they use to manually
outperformed traditional methods based on prepared features.
Studies have shown that ESA can be used for feature extraction in emotion recognition.
shows. The input layer on the convolution structure is the input of the data of the model to be trained.
is the layer that is transformed into the desired structure.
Sentiment analysis methods are divided into machine learning-based methods and dictionary-based methods.
are classified. Among these methods, considering the studies in the literature
machine learning based methods were found to give maximum accuracy.
There are some features that distinguish this project from other applications. Detailed
the fact that the success rates obtained as a result of research cannot exceed a certain limit
and it is noteworthy that most projects do not come close to this accuracy limit. This problem
In order to provide a solution, additional layers have been added to the model creation part and model training
commonly used combinations have been tested in recent research to determine certain optimal
thresholds were tried to be determined. As a result, the evaluation and success rate
has been increased. In addition, compared to other projects, we have developed a system that can return an instantaneous response value.
software has been developed and the expressions are easier to read. In this way, the model is built in a fast
and restructured and allowed to be improved.
2. PROJECT CONTENT AND SCOPE
Recently, face recognition and detection systems have been used in many commercial, military, security, social and
is frequently used in applications in psychological fields. Analyses performed; human
It involves the identification and interpretation of the movements of their faces. By humans
emotional expressions, which are difficult to analyze even in a computerized environment, can be tested and determined.
The thought that it will provide convenience has brought popularity to the field of deep learning. This
computer vision should also be mentioned in this context. Computer vision is nowadays used for face and
is widely used in the fields of emotion classification. Face recognition, image or
automatic identification or verification of persons in data derived from videos
process. There are four basic stages of face recognition. These processes are face
detection, normalization, feature extraction and classification. Normalization and
no matter how successful classification algorithms are in face recognition, if feature extraction
If this stage is not successful, the system does not achieve the desired success.
The real-time emotion detection project is a combination of image processing and artificial intelligence algorithms.
consists of a series of steps. The first step is to detect human faces in the snapshot.
face recognition algorithms were used. Once the face or faces are detected, the emotion
sentiment analysis model developed using convolutional neural networks to determine the state of
is used. This model was trained with deep learning techniques and was applied to different emotion categories.
with the allocated dataset. Finally, the identified emotional states were analyzed
and present the results to the user.
As a result of the additions made throughout the project scope, the success rate of the revamped model and
with the help of the graphs drawn that the availability increases in direct proportion.
observed. One of the deep learning techniques for feature extraction, artificial neural
a new model using Convolutional Neural Networks (ESA-CNN), an approach involving networks
has been developed.
Commonly used combinations for model training have been tested in recent studies and
The effect of the classification algorithms on the performance of the classification algorithms is analyzed.
By making evaluations, the classification algorithm with strong performance and the real
emotion classification from facial images using a time convolutional neural network architecture
The project that realizes the process simultaneously has been revealed.
3.2 Infrastructure, Hardware and Software Specifications A high quality dataset is needed for high performance of deep learning. This Therefore, we should look for datasets where training and testing performance is measured to be high and the appropriate set should be selected. In the project, the face expression recognition dataset is used for deep A project has been developed for emotion recognition with learning. Face expression recognition data set fulfills the need in the emotion detection project. Face expression recognition data There are a total of 35887 images in the set. 28821 of the images are training and 7066 of the images is reserved for Public and Private tests. Public tests are the performance after the model is finished. ratio, while Private tests are used to test a portion of the images in the dataset. "PrivateTest" and is then used for testing. Used for the project The technical details of the images can be analyzed through this data set. Thus, in columns can be seen how many groups the samples in the data set used are divided into, and the fields within the set can be called data visualization can be applied. The data set used consists of 35887 rows and 3 columns consists of. In this data set, there are pictures to identify seven emotions. This emotions angry (3993), disgust (436), fear (4103), happy (7164), sad (4938), surprise (3205), neutral (4982). The structure of the images is analyzed with the help of functions It is 48x48 in size and arranged in shades of gray. The model contains The new model was tested by training with the images separately. With the model In the study, seven different emotion classes (fear, anger, disgust, happiness, neutral, sadness, surprise) were discussed.
4.2 Areas of Use of the System-Software Facial expressions are the clearest way for people to express their inner feelings in their daily communication. is one of the indicators. A person's physical or mental state can be determined by analyzing their facial expressions. can be detected. Facial expression recognition can therefore be used in autopilot, human-computer interaction, medical treatment and other areas related to facial expressions. This field is becoming increasingly important is becoming a research topic as a result of the researches conducted. Face psychological science, such as emotion recognition from expression, identification of autism and schizophrenia, assessments, detection of a drowsy driver, Alzheimer's disease or schizophrenia identification of abnormalities in the early stages, medical inquiries, medical investigations and criminal for various applications such as prediction systems. Also facial expressions training can also be used in field applications. The use of automatic emotion recognition; digital advertising, marketing analysis, online games, customer feedback evaluation, commercial, military, security, social applications, various intelligent systems such as healthcare (e-health, learning, suggestion for tourism, smart city, smart conversation etc.) has great potential. Additional as analyzing a user's emotional reactions during interaction with products or services to evaluate the effectiveness of advertising campaigns and target audiences. Evaluate the effectiveness of advertising campaigns and target audiences can be used to develop appropriate marketing strategies. In the field of education, students can be used to monitor their emotional state and optimize the training process. In machine learning, various facial expression recognition algorithms for emotion detection has been proposed. However, the complexity, diversity, overlap, illumination problems and other difficulties in facial expression recognition, recognition in practical applications accuracy is still not satisfactory. In recent years, neural networks developments have led to effective models of deep learning and a large number of deep learning led to the development of architecture. One of these areas is facial expression recognition, computerized has an important role in vision and artificial intelligence. Deep learning, emotion recognition has shown real promise in terms of classification efficiency for problems. In particular Deep convolutional networks have revolutionized the processing of images, video and audio. For the detection, segmentation and recognition of objects and regions in images is being implemented with great success. Examples include traffic sign recognition, segmentation of biological images and natural tagged images, such as detection of faces, text, pedestrians and human bodies applications where data is abundant, autonomous robots or autonomous vehicles. CNN's The biggest recent success is facial recognition. From a person's gestures and facial expressions By examining his face, it is easy to understand what kind of emotion he has. With curiosity data usage is activated. For the benefit of users and for use in all sectors is an important study that can find a field. Considering all these benefits, sentiment analysis is both is both today's and the future's workspaces and is becoming increasingly indispensable. technology is predicted.
4.3 Potential Target Users Emotion analysis projects with human faces have been conducted in the fields of commerce, military, manufacturing, health and is interesting for service sectors. What are the companies about themselves now? They want to know what is being thought and act accordingly. In this way, deep learning-enabled systems in research areas such as economics, marketing, politics, etc. rich source of data that can be utilized. This situation has led to the development of has helped the subject to gain importance day by day. Emotion from facial expressions At the advanced stage of recognition, the concept of text mining comes to the fore. The deep learning supported emotion recognition project has a very dense target user base. has. There are many sectors mentioned in this audience. Deep learning supported emotion recognition project is integrated into camera systems, providing a rich source of data. is being collected. Processing this data source provides a wide range of information. A school system, while information about education is integrated into a military system. makes it easier to perform crime analysis through emotion detection. A If the trading company integrates the system in the store where it does its marketing, customers will have access to different products. by analyzing the emotional attitudes it creates towards brands. Brands need to be strategic, fast and help them make more informed marketing and product development decisions. This way will be ensured to demonstrate a pragmatic attitude towards its customers and internal accounting. Project if connected to a hospital camera system, it will detect the early detection of certain diseases and facilitate the identification of newly developing findings. Utilitarianism towards the school system The effect can be explained as follows: the emotional states of the students according to the courses are recorded. In this way, the attitudes of age groups towards the lessons and the difficulty of the subjects are recorded. information such as levels can be learned. In addition, it is possible to learn the international whether or not it appeals to the palate of a community of people. Determination of the state of mind of the society, film evaluations, market-price balance analysis, Scientific and medical research, Crime analysis, security, intelligence, etc., both general and potential and target users of the project include educators, psychologists, health professionals and communication specialists. Machine Interaction and Sentiment Analysis are used in marketing and advertising strategies, customer In analyzing the experience, various sub-sectors in Safety and Monitoring, Healthcare and social sciences can be used and developed in the applications of the fields. Counting all possible evaluations Unfortunately, it is not over. These systems, which are predicted to have the potential to be used in every field is not just a prediction of the future because today we are not able It is also known to be utilized.
