Table 1 Summery of the related work.
From: Improved facial emotion recognition model based on a novel deep convolutional structure
Authors | Year | Publisher | Pros | Limitations & Challenges |
|---|---|---|---|---|
Umer et al.3 | 2022 | Springer | The trade-off between augmentation and deep features has an impact on the detection ability of the FER systems in unfamiliar test samples. | - |
Zang5 | 2019 | ICML conf. | Avoids aliasing issues caused by down sampling, demonstrating that blur filters perform effectively for an extensive variety of visual identification applications. | Efficiency depends on type of CNN architectures and the recognition task. |
Chowdary et al.9 | 2021 | Springer | Achieved higher performance after eliminating the fully connected layers from pre trained models. | The proposed model was using only one dataset in testing phase. |
Abate et al.10 | 2022 | Springer | Investigated the influence of masked faces on recognizing emotions pointing attention to the challenging occlusion problem. | Low FER accuracy which still needs to be improved. |
Shaik et al.11 | 2022 | Springer | A deep neural network was constructed that accepts both local and global attention information that outperformed many recent advances in FER. | Only operated on frontal pictures and was confined to real-time invariant face data. |
Saurav et al.12 | 2022 | Springer | Outperformed current CNN models in terms of computing efficiency and recognition accuracy. | Performed poorly in fear class. |
Rajan et al.13 | 2020 | IET Image Processing | Performed well in distinguishing surprised and joyful reactions. | Mis-classification in sadness and anger. |
Khattak et al.14 | 2022 | Springer | An effective method to extract age, gender, and emotions information from facial images. | Just one dataset was carried-out on gender and age experiments. |
Bentomi et al.15 | 2022 | Springer | avoid overfitting by using the early stopping criterion and also improved the overall accuracy. | The small size of the datasets |
Liu et al.16 | 2021 | IEEE | Utilization of face alignment to minimize the influence of ambient noise. Efficiently distinguishes between comparable sentiments such as fear and disgust. | The accuracy still needs to be improved. |
Wang et al.17 | 2021 | Wiley Online Library | Solve the problem of two tasks divergence (classification and regression). | Classification task and emotion recognition accuracy still need to be improved. |
Taskiran et al.18 | 2020 | Wiley | Useful for performing face recognition in videos extracted from systems that may contain images with illumination variations, noise, and blur while performing face recognition | The accuracy still needs to be improved for better face recognition. |
Saurav et al.19 | 2021 | Springer | The suggested model functioned well in identifying facial images in the neutral, surprise, disgust, and happiness classes | Struggled in the sad and afraid classes |
Devi et al.20 | 2021 | Springer | achieved significant classification accuracy | The major issue was the high training time |
Li et al.21 | 2021 | ScienceDirect | Overcome the overfitting problem that may occur during training phase | More images were needed to be collected to propose a better-optimized algorithm and also to make more improvements in facial recognition |
Arora et al.22 | 2021 | Springer | Achieved high classification accuracy. | Only one dataset in the testing phase was used |
Zheng et al.23 | 2022 | IEEE | Increase the capacity of instructors to recognize expressions in real-world world environments. A new dataset of intensity-based facial expressions known as EIDB-13 was generated. | For better feature extraction, the proposed approach has to be optimized further. |
Fontaine et al.22 | 2022 | Wiley Online Library | The model was beneficial in recognizing severe pain, and accurately predicted pain intensity. | The scientists did not compare the results to human observers’ assessments. |
Lu et al.23 | 2022 | ScienceDirect | Well recognized and identified the face and emotion in real time. | Limited to three emotions only (happy, regular, and unhappy). |
Mohan et al.6 | 2020 | IEEE | Succeeded to extract local characteristics from face images. | The high training time of this work hindered its performance |
Mohan et al.26 | 2021 | Springer | Simple architecture that can recognize and identify the face and emotion in real time. | Average recognition rates on all applied datasets. |
Mohan et al.27 | 2021 | IEEE | DCN model which used to accurately identify the multiscale variations of deceit. | - |
Suzuki et al.28 | 2022 | IEEE | Applying the anti-aliased CNN in data-limited situations achieving more accurate results. | Need to test model on more datasets for generalization. |
