Further Development of Advanced AI Facial Recognition Technology

2020-01-02

The application of artificial intelligence in facial recognition technology seems to be one of the fastest developing technologies to date. According to ZDNet, companies like Microsoft have developed facial recognition technology that can use emotion tools to identify facial expressions. However, there are still limitations to these tools, as they are restricted by eight so-called core states: anger, contempt, fear, disgust, happiness, sadness, surprise, or neutrality. Japanese technology developer Fujitsu has introduced an artificial intelligence-based technology that takes facial recognition a step further in tracking emotional expressions. The existing facial recognition technology is based on "recognizing various action units - that is, certain facial muscle movements we make, which are associated with specific emotions." The problem with current technology is that artificial intelligence needs to be trained on a large dataset. It needs to know how to recognize from all possible angles and positions. Without enough images, it is not as accurate in typical situations. Training artificial intelligence with a large amount of data to effectively detect emotions is very challenging. For example, if the subjects in the experiment do not sit in front of the camera and look directly at the lens as required, the experiment becomes extremely difficult, and there are many similar problems. Fujitsu claims to have found a method to improve the quality of facial recognition results in emotion detection. Their newly developed tool's task is to "extract more data from a single image" instead of using a large number of images to train artificial intelligence. The company refers to this process as a "standardization process," which converts photos taken from specific angles into images similar to front-facing shots. With the same limited dataset, it can better detect more action units, even from a tilted angle, and identify complex emotions that are more subtle than the core expressions analyzed by current methods. This tool can detect complex emotional changes, such as nervous laughter, with an accuracy rate of 81%, determined through "standard evaluation methods." In comparison, independent research statistics show that Microsoft's tool has only a 60% accuracy rate, which decreases when dealing with photos taken from more tilted angles.

RECENT AI NEWS

RECENT AI TOOLS