IMPROVED ANALYSIS MODEL FOR FACIAL EXPRESSION DETECTION FROM LOW-RESOLUTION IMAGES

Abstract

Facial Expression Detection and the classification of images according to the recognized human emotion has its own significance in the development of the latest artificial intelligence-based systems and the latest research trends. The new technologies and facial expression detection systems are playing a vital role in improving the accuracy of results. However, existing systems perform better when working on high resolution image. The results of Facial expression detection for human emotion recognition from low-resolution images are still uncertain. In this research, an artificial intelligence evaluation model is developed to classify and evaluate the performance of facial expression detection systems for low-resolution images. The performance of three facial expression detection systems, including Azure Face API, Face++ and Face Reader, is analysed in two facets: (1) human emotion recognition from low resolution image, and (ii) the classification of these images according to the class of human emotion through the trained model. The training is performed on the dataset to evaluate the performance in terms of precision, recall, mAP and f-score. These three systems also perform accurately with high-resolution images. The Azure Face API provide more accurate results than the other two systems (i.e., 80% of the images are manipulated and classified accurately). However, there are some limitations for some specific human facial expressions. So, there is a need for experimental evaluation of these systems for the guidance of future system developments and research in the discipline of facial expression detection.