Application of expression recognition based on Genetic Algorithm in English distance teaching platform

doi:10.21203/rs.3.rs-2716041/v1

Download PDF

Research Article

Application of expression recognition based on Genetic Algorithm in English distance teaching platform

https://doi.org/10.21203/rs.3.rs-2716041/v1

This work is licensed under a CC BY 4.0 License

Journal Publication

published 23 May, 2023

Read the published version in Soft Computing →

You are reading this latest preprint version

In Distance English teaching, teachers cannot take care of every student because of their own limitations. Therefore, more advanced technology is needed to assist teachers in completing teaching. Based on this point, this paper proposes a distance English teaching assistant method based on facial expression recognition technology. First, the distance learning platform collects the image data of students through the camera, preprocesses the captured image, detects and recognizes the skin color, and finally determines the position of the face. Then, based on the theory of genetic algorithm, a model to improve the characteristics of face images is established, and the recognized facial expression images are classified by random forest algorithm, and the facial expression characteristics of students are analyzed according to the five kinds of information of the target, such as attention, suspicion, distraction, excitement and anxiety. This paper adds the design of expression recognition application to the English learning distance platform, collects and summarizes the changes of students' expressions through the mobile terminal, and reports them to the teacher as reference information in the teaching process. The results show that the facial expression recognition module carried on the distance English teaching platform has a relatively accurate recognition rate, can provide some help for English teachers, and is well received by teachers, so it is worth promoting. By studying genetic algorithm and facial expression recognition technology, this paper applies them to the design of English distance teaching platform, which has a certain auxiliary effect on English teachers' teaching behavior.

Genetic algorithm

Expression recognition

English teaching

Remote platform

Due to the rapid progress of information technology, people's learning and lifestyle have also undergone major changes. Education for all, high-quality education, lifelong learning, collaborative learning and personalized learning are gradually becoming the main themes of the information technology era, and distance education is also gradually developing [1–2]. However, in real life, the "teaching" and "learning" of Networked Distance platform learning do not have rich situational communication and communication. For example, teachers and students can't see each other's expressions, expressions, gestures, etc., so teachers can't understand students' learning more accurately, which will lead to the loss of the advantages of traditional teaching [3]. When students are facing learning difficulties or challenges, they are difficult to get the answers they want, and they cannot solve their inner frustration. In the face of cold electronic equipment and machines for a long time, they will gradually lose interest and confidence in learning [4]. Therefore, the method to solve the lack of sensory communication in distance learning system has become the main problem of Networked Distance Learning. In interpersonal communication, facial expression has always played an important role. In fact, facial expression is not only a way for people to express their emotions, but also a means of interpersonal communication [5–6]. Through these precise and subtle facial expressions, people can not only express their thoughts and feelings, but also recognize each other's emotions and inner world through expressions. Facial expressions contain a lot of information about the communication behavior between people. The research on them can better understand people's real-time psychological state [7]. Expression analysis is a process of automatically, reliably and efficiently obtaining the information of facial expression communication. In the past decade, with the continuous development of expression recognition technology, it has greatly promoted the development of human-computer interaction and image understanding technology, and contributed to the popularity of intelligent computers [8]. Therefore, the research of facial expression is very important. Inspired by various related applications, facial expression recognition technology has made great progress.

The literature shows that distance education goes beyond the boundaries of time and space. People can use mobile devices to carry out distance learning on the network without worrying about the impact of time and space. Distance learning can give full play to students' subjective initiative and spontaneity, and can fully study. With the advancement of the times and the diversification of people's lifestyle, distance education has been favored by many people, and its importance and self-worth have also been recognized by many people [9–10]. The literature believes that networked distance learning can subvert the one-way teaching mode of "teacher led". Teaching and learning is an interactive process. The traditional teaching mode is often in the state of "teacher led, students follow", which will make students lose the right to speak. The right to speak is the core of the two-way teaching process. Students in real life prefer more direct and convenient communication and teaching feedback [11]. Literature shows that distance education can stimulate students' interest in learning and increase students' autonomy to participate in learning. Compared with paper-based teaching plans, distance learning through network distance technology can provide a teaching platform with pictures and texts, and can design rich and interesting teaching forms, such as courses and teaching content teaching plans, which can greatly improve students' interest and enthusiasm in learning, as well as their autonomy [12]. The literature shows that the use of remote equipment for auxiliary teaching can improve students' learning enthusiasm and learning efficiency. The literature believes that genetic algorithm can simulate natural selection in nature, and can design and calculate according to Darwin's biological evolution theory and genetic mechanism [13]. This technology has become the main core technology of intelligent computers at present. The literature shows that in the early stage of population development, a small amount of selection pressure is usually used to maintain population diversity; With the progress of evolution, in the late stage of evolution, in order to improve the competition of excellent individuals and find the optimal solution of the development direction, greater selection pressure is often used [14]. In the process of distance learning, the literature divides the research content of emotion calculus into nine aspects: emotion mechanism, emotion information acquisition, emotion recognition, Emotion Modeling and understanding, emotion comprehensive expression, emotion computing application, emotion and computer interface, emotion transmission and communication, wearable computer, etc. The literature divides emotions in human physiology into "rough" and "fine". "Rough" refers to the rapid response of all parts of the body, such as anger, enthusiasm, etc; "Fine" reactions are not so strong, such as calm. According to the literature, expression is the characteristic of external expression of emotion, and we can express and judge human emotions through expression [15]. Among human emotional expressions related to posture and sound, facial expressions contain the most emotional information, can best recognize emotions with different characteristics, and can express information that cannot be expressed in language or sound. Therefore, in the process of interpersonal communication, facial expression plays an important role and is the main symbol of emotion recognition. According to the literature, facial expression recognition is an important way to obtain emotional information in the process of human-computer emotional interaction. In the real world, such as job hunting and other application scenarios, facial expression information is easier to retrieve than other emotional information, and can be retrieved directly from cameras and other devices in real time [16]. The literature divides the human face into several small activity units, each of which is called activity unit (AU). An Au may contain one or more muscle tissues. Ekman summarized the relationship between facial activity units and expression changes through a large number of studies, and pointed out that most facial expressions are composed of several activity units (AU). These combined activity units are interactive and have the difference between primary and secondary [17]. If an activity unit is dominant in facial expression, the characteristics of secondary activity units in facial expression are not clear and can be ignored. Literature shows that in classroom teaching, students' expression is a form of feedback to the current content understanding, and it is also a common form. When students understand the current knowledge points, their expression is usually happy and confident; When students are difficult to understand what they are learning now, their expressions are often anxious and confused; If students are tired of learning or are not interested in what they have learned, they usually appear distracted [18].

3.1 Image preprocessing

The normalization of histogram is to expand the gray level of the original image from N_min to N_max. In this study, N_min =0, N_max =255. The input light level is shown by formula (1):

$${N}_{x,y}=\frac{{N}_{max}-{N}_{min}}{{o}_{max}-{O}_{min}}-\left({O}_{x,y}-{O}_{min}\right)+{N}_{min}\forall x,y\in 1,N$$

The algorithm of histogram equalization is as follows:

Histogram equalization is usually divided into three steps: first, count the number of occurrences of each gray level in the original image, second, accumulate the normalized histogram, and finally calculate new pixels.

Therefore, the number of pixels in each brightness level will not change after the image is leveled.

$$\sum _{l=0}^{M} O\left(l\right)=\sum _{l=0}^{M} N\left(l\right)$$

M represents the number of light levels in the image, O (l) represents the number of l-level pixels, and N (l) represents the number of L-level pixels after equalization.

$$\sum _{l=0}^{p} O\left(l\right)=\sum _{l=0}^{q} N\left(l\right)$$

$$N\left(l\right)=\frac{{N}^{2}}{{N}_{max}-{N}_{min}}$$

The cumulative histogram after histogram equalization is:

$$\sum _{l=0}^{q} N\left(l\right)=q\times \frac{{N}^{2}}{{N}_{max}-{N}_{min}}$$

Bring formula (3) into the above formula

$$q\times \frac{{N}^{2}}{{N}_{max}-{N}_{min}}=\sum _{l=0}^{p} O\left(l\right)$$

After conversion, the output pixel mapping of brightness level q can be obtained:

$$q=\frac{{N}_{max}-{N}_{min}}{{N}^{2}}\times \sum _{l=0}^{p} O\left(l\right)$$

E represents the equalization function of image 0, then the output image is:

$$\begin{array}{c}E(q,O)=\frac{{N}_{max}-{N}_{min}}{{N}^{2}}\times \sum _{l=0}^{p} O\left(l\right)\\ \end{array}$$

$${N}_{x,y}=E\left({O}_{x,y},O\right)$$

After the image and histogram are equalized, it can be clearly seen that the surface becomes clearer, and the dynamic range of gray scale can also be significantly improved in the histogram.

3.2 Skin color detection and recognition

Skin color is an important intuitive feature of human face. Skin color reflects a set of color space. Face recognition methods based on skin color often use the information of image color, which can greatly improve the detection speed.

The color space used in this paper is YCrCb color space. In the previous step, the image processing reduces the detection interval and reduces the number of calculations. The image generated by computer is RGB model, and YCbCr model is obtained by linear conversion to RGB format. The conversion formula is shown in (10):

$$\left[\begin{array}{c}Y\\ {\text{C}}_{\text{b}}\\ {\text{C}}_{\text{r}}\end{array}\right]=\left[\begin{array}{c}16\\ 128\\ 128\end{array}\right]+\left[\begin{array}{ccc}65.481& 128.553& 24.966\\ -37.797& -74.203& 112.000\\ 112.000& -93.786& -18.214\end{array}\right]\left[\begin{array}{c}R\\ G\\ B\end{array}\right]$$

According to the distance from the point to the center point, we can get the uniformity of skin color. The gray value of each pixel is equal to the pixel and color of the skin. The formula is as follows:

$$\text{p}(\text{C}\text{b},\text{C}\text{r})=\text{e}\text{x}\text{p}[-0.5](\text{x}-\text{m}{)}^{\text{T}}{\text{C}}^{-1}(\text{x}-\text{m})$$

The image is segmented by using the method of fixed threshold. The threshold is set to 0.6. The area with gray value greater than 0.6 on the probability map is skin color area, otherwise it is non skin color area. So far, we have obtained the candidate region of human face.

3.3 Facial position determination

The steps of AdaBoost algorithm are as follows:

Order (x1, Y1), (xn, yn) is the training sample set, where xi ∈ X = {0,1}, yi ∈ Y = {0,1}

Starting weight

$${\omega }_{1,i}=\left\{\begin{array}{c}\frac{1}{2m},{y}_{i}=0\\ \frac{1}{2l},{y}_{i}=1\end{array}\right.$$

Where m is the number of positive samples; l is the number of counterexample samples.

Fort = 1,.,T

I: weight normalization

$${\omega }_{(t,i)}\leftarrow {\omega }_{(t,i)}/\sum _{j=1}^{n} {\omega }_{t,j}$$

II. Train feature classifier hj for each feature j, and calculate the and error rate of added weight ωt

$${\epsilon }_{\text{j}}=\sum {\omega }_{\text{i}}\left|{\text{h}}_{\text{j}}\left({\text{x}}_{\text{i}}-{\text{y}}_{\text{i}}\right)\right|$$

ⅲ: select the weak classifier HT with the lowest error rate.

Ⅳ: update the weight.

Finally, a strong classifier is obtained

$$h\left(x\right)=\left\{\begin{array}{cc}1,& \sum _{t=1}^{r} {\alpha }_{t}{h}_{t}\left(x\right)\ge 1/2\sum _{t=1}^{r} {\alpha }_{t}\\ 0,& \text{ others }\end{array} {\propto }_{t}=\text{l}\text{o}\text{g}\frac{1}{{\beta }_{t}}\right.$$

We can use the trained classifier H (x) to detect the face image and get the correct position of the face. In this step, a specific area of the face is obtained.

4.1 Theoretical basis of genetic algorithm

The flow of genetic algorithm is shown in Fig. 1.

The linear combination of any two points x1 and x2 of the arithmetic crossover operator in the solution space d of the convex search space is:

$$p{x}_{1}+(1-p){x}_{2}\text{, }p\in \left[\text{0,1}\right]$$

Assuming that x1 and x2 in the population are selected as the two parents of the crossover operation, two parents produce children as follows:

$$\begin{array}{c}{x}_{1}{ }^{{\prime }}=p{x}_{2}+(1-p){x}_{1}\\ \end{array}$$

$${x}_{2}^{{\prime }}=p{x}_{1}+(1-p){x}_{2}$$

Replacing the original gene value and changing it into a random number is the uniform mutation operation. This random number will be evenly distributed and located in the sequence of the corresponding gene value. Therefore, individuals can act independently in the search space, but they are not good at the local search process for specific values. In order to make up for this deficiency, we will make a small random change to the value of the original gene, take the modified value as the new value of the gene, and replace the original value of the gene with the new gene value, rather than taking the same random number distribution. After the mutation operation with equal probability on each trajectory, the space corresponding to the solution changes slightly along the solution vector.

$$\begin{array}{c}{X}^{{\prime }}=\left({X}_{1},{X}_{2},\cdots ,{X}_{k}^{{\prime }},{X}_{1}\right)\\ {X}_{k}^{{\prime }}={X}_{k}+\varDelta X\text{ }\text{o}\text{r}\text{ }{X}_{k}-\varDelta X\end{array}$$

If there is a parent x, if the element xk is selected to mutate, the result of mutation is: where Δ X is the changed value.

4.2 Facial image feature enhancement

For an n × n given image of pixels is first divided into m × m personal face feature block. Each face shape block is actually a 1 × 1 pixel image. Define the initial population: a_ij is the block state of the face shape. If the block on the surface is significant, set ia_j to 1, otherwise set to 0. All surface features of a_ij block are composed of matrix s_k, which is as follows:

$${S}_{k}=\left[\begin{array}{cccc}{a}_{11}& {a}_{12}& \cdots & {a}_{1m}\\ {a}_{21}& {a}_{22}& \cdots & {a}_{2m}\\ ⋮& ⋮& \ddots & ⋮\\ {a}_{m1}& {a}_{m2}& \cdots & {a}_{mm}\end{array}\right],k=\{\text{1,2},3,\dots N\}$$

Among them, SK is one of the possible solutions of the optimal part of the expression of a specific population, and N is the population size. Calculation parameters α: Let a training image of the same training expression in the sample set be D, then take the face feature matrix of the training image D, and take the matrix as the I_D. If the following rules are met, the training image D belongs to this kind of expression:

$$\left|{I}_{D}\cap {S}_{k}\right|\ge {\Omega }\sum _{r=1}^{m} \sum _{c=1}^{m} {a}_{rc}$$

Ω is the threshold set to 0.8, which means that it must have 80% similarity with the selected feature to indicate that it belongs to this kind of expression.

4.3 Facial expression recognition

For a given image sample A, image features are extracted through the optimal mapping vector, so that

$${\text{y}}_{\text{k}}=A{\text{x}}_{\text{k}},\text{k}=\text{1,2},\cdots ,\text{d}$$

For a single network, we use hidden layer RBF network. The output layer of each network has an output result representing the expression category, and the size of the image sample feature matrix calculated above is m × d.

Among them, when using genetic algorithm to select a single neural network for integration, it is assumed that M neural networks N₁, N₂, N₃, Nm is trained independently, and N simple averaging methods are used to form a neural integration network. After considering the elimination of nm neural network, N'neural network is composed of N₁, N₂, N₃, N_m−1 uses a simple average method to meet:

The correlation between neural network N_i and N_j is defined as follows:

$${C}_{ij}=\int p\left(x\right)\left[{N}_{i}\left(x\right)-d\left(x\right)\right]\left[{N}_{j}\left(x\right)-d\left(x\right)\right]dx$$

If a genetic individual corresponds to its own s in {N₁, N₂, N₃,..., n_m}, assuming that the verification set is V, the estimated value of the correlation between Ni and NJ in the calculated neural network verification set V is:

$${C}_{ij}^{V}=\sum _{x\in V} \left[{N}_{i}\left(x\right)-d\left(x\right)\right]\left[{N}_{j}\left(x\right)-d\left(x\right)\right]/\left|V\right|$$

The average error of the neural network corresponding to s integrated in the verification set V is:

$$\left(\sum _{{N}_{i},{N}_{j}\in S} {C}_{ij}^{V}\right)/|S{|}^{2}$$

The fitness value of the genetic algorithm is selected as the reciprocal of the error.

4.4 Classification of facial expression recognition results

After selecting facial features, they are divided into five expressions: focus, doubt, distraction, excitement and anxiety. Then a random forest classification algorithm is proposed, which can effectively improve the recognition accuracy of expression classification.

In this algorithm, the Gini index is the standard for judging the random forest regression tree, and the calculation process is:

$$\text{G}\text{i}\text{n}\text{i}\left(S\right)=1-\sum _{i=1}^{mtry} {P}_{i}^{2}$$

Where P_i represents the probability that Y_i class will appear in test set S.

The schematic diagram of two-dimensional facial expression recognition algorithm based on feature extraction and random forest feature classification is shown in Fig. 2:

4.5 Facial expression recognition and classification result analysis

There are two problems to be solved in the experiment: the selection of the number of hidden layer nodes and the number of subnets in RBF network. Firstly, the selection of the number of hidden layer nodes in RBF neural network is determined. We set the number of hidden layer nodes to 2, 5, 10, 15, 20, 25 and 30 for experiments, and use RBF neural network to recognize facial expressions. The results show that if the number of hidden layer nodes starts from 2, the detection rate increases with the increase of the number of nodes. If the number of nodes is 15, the detection effect is the highest, and then the detection rate decreases with the increase of the number. The experimental results of hidden layer node parameters comparison and debugging are as follows:

According to the experimental results, the number of hidden layer nodes is set to 15. The number of hidden layer nodes in RBF neural network is determined. Then determine the number of subnets. The experimental results are:

In the previous step, it has been proved that if the number of networks is 6, the detection rate is the best. In order to use genetic algorithm to select individuals from networks with multiple differences, we trained 6 * 2 individual networks, and then genetic algorithm is used to select 6 * 2 individuals from the network. Select 6 of them for network integration.

We also compare other algorithms. In the experiment, three different facial expression recognition algorithms are provided. Most of these algorithms have achieved high accuracy. All these algorithms and their accuracy in five facial expression recognition are shown in Table 1. It can be seen from Table 1 that the algorithm designed in this paper has achieved the best results.

Table 1

Comparison of accuracy of different face recognition algorithms
Method Expression	Algorithm A	Algorithm B	Algorithm C	Algorithm D	Algorithm E	Algorithm in the paper
Focus	0.4	0.6	0.7	0.6	0.94	0.980
Doubt	0.7	0.9	0.65	0.94	0.85	0.94
Distracted	0.7	0.6	0.8	0.9	0.9	0.980
Excited	0.86	0.85	0.87	0.90	0.9	1
Anxiety	0.7	0.76	0.8	0.9	0.82	0.95
Average	0.67	0.75	0.75	0.84	0.88	0.97

5.1 Expression recognition application design

The flow chart of English distance learning platform system is shown in Fig. 5:

The topology of English distance learning platform based on facial expression recognition is shown in Fig. 6. Users can connect to the network through WiFi to realize learning information feedback.

After opening the platform, the face recognition system runs in the background, and automatically collects the facial expression photos of students in the learning process and uploads them to the server. The server recognizes the state of facial expression and gives feedback.

5.2 Statistical methods of expression characteristics

Attention level refers to students' listening state in class; The degree of participation refers to the degree of students' participation in classroom teaching and the embodiment of students' dominant position in the classroom; Difficulty refers to students' ability to understand the current teaching content; Active time is the active state of students in classroom teaching. This study calculates the general attention, participation and doubt difficulty in the classroom from the perspective of emotion, so as to help teachers better measure the quality of classroom teaching. The calculation of students' attention, participation and difficulty is obtained by analyzing the images divided into t time intervals.

5.3 Application process of expression recognition system

In practical applications, you can first retrieve video data from camera devices or local files. Then, by setting the time and frequency, the system automatically collects images or manually collects images, carries out facial analysis and partial image acquisition of expression, then uses random forest algorithm to classify expression, and finally stores the expression classification results and the corresponding time in the database. Among them, the construction of feature library needs to use a large number of expression images to capture expression features and save them in the feature library based on people's previous understanding of expression. Feature library is the basis of expression recognition.

The intelligent emotion evaluation system has the function of automatic expression recognition. Before the class, the main body of the system starts first, and then enters the circular program to carry out face recognition, feature extraction, expression classification in turn, and store the expression recognition results. Teachers can set an appropriate time interval as needed to recognize facial expressions in the image every T seconds.

The recognition expression evaluation system also has the function of artificial expression recognition, and the system will follow the instructions of the teacher. This function helps teachers pay attention to the listening situation of individual students and helps to improve the effect of classroom teaching.

5.4 Application effect of expression recognition system

Figure 7 shows the collected attention comparison results every five minutes from 14:00 to 15:00.

After the experiment, talk with the teachers in the form of interview to understand their feedback on the application of expression recognition and evaluation system. Through the teacher's feedback, we learned that the intelligent emotion evaluation system is easy to use and has high accuracy of recognition results, which is helpful to improve the level of classroom teaching and has great practical value for traditional classroom teaching.

The intelligent system of emotional teaching evaluation is the product of the era of educational informatization. Its emergence can promote the development of intelligent teaching methods. But in the research process, we found that the system still has many shortcomings that need to be improved. For example, the system uses facial expression recognition technology to evaluate students' learning status, so it has high requirements for facial expression recognition technology, and the recognition process will be affected by many factors. Therefore, it is necessary to further improve the classroom facial expression recognition algorithm to make the recognition results more accurate and faster.

Conflict of interest

The authors declare that they have no conflict of interests

Ethical approval

This article does not contain any studies with human participants performed by any of the authors.

Data Availability

Data will be made available on request.

B. Williamson, Big Data in Education: The Digital Future of Learning, Policy and Practice, Sage, Thousand Oaks, CL, USA, 2017.
S. Tang, “Learning mechanism and function characteristics of MOOC in the process of higher education,” Eurasia Journal of Mathematics, Science and Technology Education, vol. 13, no. 12, pp. 8067–8072, 2017.
A. K. Milicevic, B. Vesin, M. Ivanovic et al., “E-learning personalization based on hybrid recommendation strategy and learning style identification,” Computers and Education, vol. 56, no. 3, pp. 885–899, 2011.
J. Jo, H. Jun, and H. Lim, “A comparative study on gamification of the flipped classroom in engineering education to enhance the effects of learning,” Computer Applications in Engineering Education, vol. 26, no. 5, pp. 1626–1640, 2018.
S. Li and W. Deng, “Deep facial expression recognition: a survey,” IEEE Transactions on Affective Computing, 2020, https://arxiv.org/pdf/1804.08348.pdf.
S. Zhao, H. Cai, H. Liu, J. Zhang, and S. Chen, “Feature selection mechanism in cnns for facial expression recognition,” BMVC, p. 317, 2018.
A. Yao, D. Cai, P. Hu et al., “Holonet: towards robust emotion recognition in the wild,” in Proceedings of the 18th ACM International Conference on Multimodal Interaction, pp. 472–478, Tokyo, Japan, November 2016.
X. Yao and Zhao, “Facial expression recognition based on convolutional neural network local feature fusion,” Program Laser Optoelectronics, vol. 57, pp. 338–345, 2020.
B. L. Cai, Q. Lin, G. Liang, Y. L. Shi, and H. C. Sun, “Study on the transformation of college students’ learning style in online teaching environment: problems, trends and development paths,” Heilongjiang Researches on Higher Education, vol. 12, pp. 140–144, 2020.
R. R. Sinkovics, P. Haghirian, and S. Yu, “Information technology‐based innovation in international marketing education: an exploration of two learning environments,” Journal of Teaching in International Business, vol. 20, no. 2, pp. 123–148, 2009.
A. Segal, Z. Katzir, B. Shapira et al., “Edurank: a collaborative filtering approach to personalization in e-learning,” in Proccedings of the 7th International Conference on Educational Data Mining, pp. 68–75, London, UK,, July 2014.
D. Gamage, I. Perera, and S. Fernando, “MOOCs lack interactivity and collaborativeness: evaluating MOOC platforms,” International Journal of Engineering Pedagogy (iJEP), vol. 10, no. 2, pp. 94–111, 2020.
A. Juskeviciene, E. Jasute, E. Kurilovas, and J. Mamcenko, “Application of 1: 1 mobile learning scenarios in computer engineering education,” International Journal of Engineering Education, vol. 32, no. 3, pp. 1087–1096, 2016.
F. Zhang, T. Zhang, Q. Mao, and C. Xu, “Joint pose and expression modeling for facial expression recognition,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 3359–3368, Salt City, UT, USA, June 2018.
Z. He and Chen, “Convolution neural network with multi-resolution feature fusion for facial expression recognition,” Laser Optoelectronics Program, vol. 55, pp. 370–375, 2018.
W. Pei, H. Dibekliog˘lu, T. Baltrušaitis, and D. M. Tax, “Attended end-to-end architecture for age estimation from facial expression videos,” IEEE Transactions on Image Process, vol. 29, pp. 1972–1984, 2019.
X. Feng, M. Pietikainen, and A. Hadid, “Facial expression recognition based on local binary patterns,” Pattern Recognition and Image Analysis, vol. 17, no. 4, pp. 592–598, 2007.
J. Cai, Z. Meng, Z. Li et al., “Island loss for learning discriminative features in facial expression recognition,” in Proceedings of the 2018 13th IEEE International Conference on Automatic Face & Gesture Recognition (FG 2018), pp. 302–309, IEEE, Xian, China, May 2018.

Download PDF

Journal Publication

published 23 May, 2023

Read the published version in Soft Computing →

Reviewers agreed at journal
27 Mar, 2023
Editor assigned by journal
25 Mar, 2023
First submitted to journal
20 Mar, 2023

You are reading this latest preprint version

Application of expression recognition based on Genetic Algorithm in English distance teaching platform

Status:

Journal Publication

Version 1

Abstract

Figures

1. Introduction

2. Related Work

3. Face Image Extraction And Recognition

3.1 Image preprocessing

3.2 Skin color detection and recognition

3.3 Facial position determination

4. Expression Recognition Based On Genetic Algorithm

4.1 Theoretical basis of genetic algorithm

4.2 Facial image feature enhancement

4.3 Facial expression recognition

4.4 Classification of facial expression recognition results

4.5 Facial expression recognition and classification result analysis

5. Design And Application Of Expression Recognition In English Distance Teaching Platform

5.1 Expression recognition application design

5.2 Statistical methods of expression characteristics

5.3 Application process of expression recognition system

5.4 Application effect of expression recognition system

6. Conclusion

Declarations

References

Status:

Journal Publication

Version 1