Products Appreciation by Facial Expressions Analysis

摘要

Automatic assessment of users' appreciation of products represents an important functionality for shops, leading to better fitted products on the customers' needs and enabling more efficient marketing strategies. By means of a surveillance system we track customers, make a first interpretation of their behaviour and analyze their facial expressions when they are next to a product. Facial expressions carry relevant information regarding customers' opinion of products and can be used to detect if they show interest and also which type (positive or negative). The main contribution of this work resides in the development of a facial expression recognition analyzer that can be used in the product appreciation domain. In our approach we employ the Active Appearance Model to extract the key facial regions (e.g. eyes, nose, and mouth). Around these special regions we define Regions of Interest and extract relevant features using the optical flow estimation method. The classification phase is carried out using Hidden Markov Models. Experiments are conducted on the well-known Cohn-Kanade database and also on our own recorded database of 21 product emotions to show the efficacy of our approach. An average recognition accuracy of 93% is achieved.