Study of Smile Detection Methods

Abstract

A happy face or smile is a common expression in our daily life. This brings out basic hidden emotions, such as satisfaction and happiness. It is the most powerful and challenging tasks in social communication. Smile detection is a feature of a digital camera that resists the image from being captured until and unless person is smiling. Smile detection is divided into two parts or phases. Firstly, it detects a face from image or video, then waits for smile. In more detail we can say, a motion detector splits the image or video into frames, then analyzes frameworks such as flash and balanced level for the facial region. When a person laughs or smile, the camera detects a deformation by identifying multiple criteria, such as raised cheeks, upturned mouth visible teeth, and narrowed eyes, etc. Smile detection is used in many applications such as automatic image capturing, interactive systems, video conferencing, patient monitoring, product rating, etc. Effective representation of smile is important for smile detection and image retrieval applications. In this report we have studied various smile detection methods and proposed two smile-detection models. In the first model we use LBP and SOM classifier. The second model is extended version of first models. The second model has two consecutive actions: 1) amalgamation of Geometric Feature Extraction (GFE) and regional Local Binary Pattern (LBP) features extraction using autoencoders; 2) Self-Organizing Map (SOM) is adopted to classify smile based on these features. A comprehensive evaluation of the proposed models on a benchmark dataset GENKI-4K for smile detection shows a notable improvement in terms of performance measures on respective datasets as compared to the other popularly used smile detection methods.