The facial emotional state classification problem belongs to the class of data mining problems from images and videos. The input of the problem is a data set of images, images containing faces or video data. The facial emotional state classification problem will classify facial states into categories such as: happy, happy, surprised, scared, disgusted, angry (see figure 1). This is a problem that can be applied in a number of fields such as intelligent education (analysing students’ emotional states while listening to lectures,…), providing services/products to customers, and smart healthcare. , behavior analysis (combined with problems such as gesture recognition, speech, …) or human-machine interaction.
image 1. Example of face states in ITI data set
- Data preprocessing: return face image to standard form; If the input is an image containing a face or a video, a face feature extraction and detection process is required.
- Face classification: can use supervised learning techniques (Deep learning, SVM, Decision Tree, Random Forest, Bayes,…), semi-supervised learning (Semi-supervised learning), or unsupervised learning (K) -Means, DBSCAN,…).
- User interaction: Allows the user to have feedback then the system will integrate the feedback to increase the quality of the system.
- Performance and simulation of results
- The problem of feature extraction, image acquisition: The results of the classification system are often affected by the quality of the data. For image data, the extraction of features is a decisive factor, but the image is often noisy by many factors such as brightness, tilt angle, acquisition means, … So this step requires There are many different tests to choose the right features.
- The problem of choosing machine learning techniques: Many machine learning techniques are being researched and developed, but choosing the right method for each problem is not simple, even for each technique or problem. how to choose the right set of parameters on a data set is also a time consuming job. Research might focus here including developing/improving new machine learning techniques or building hybrid methods with other techniques to increase the effectiveness of machine learning techniques.
- The problem of integrating feedback: In the process of using the system, there are Q&A sessions with users to find more suggestions and then integrate the feedback into the classification process to get results. better.
- Speed problem: Many problems need to be handled with big data, the problem of applying big data techniques also needs to be studied.
- Integration with physical devices: the system can integrate with devices such as cameras, webcams to perform online analysis, integrate on other mobile applications as part of the system, or build systems that combine both hardware and software in specialized devices.
