Detection of Kinship through Microexpression Using Colour Features and Extreme Learning Machine

Abstract

Kinship detection based on human face image is something new and quite challenging problem in computer vision pattern recognition. There have been many applications developed to analyze social media and adopted children. Most of the existing kinship methods assume that each pair of images with a positive facial image (with an image confirming kinship) has a greater score for the group of non-negative kinship images. In practice, however, these assumptions are usually over-sampled from real-life palettes. In this research, activity involving microexpression is used as an alternative to the other parameters in kinship detection. Using several reference methods such as color features and extreme learning machines as feature extraction and classification. This set of methods offers the advantages in identifying kinship relationship. Microexpression parameters are widely used as a reference. ELM itself has the advantage of extensive training time and faster performance than other methods. The purpose of this study is to maximize classification performance and minimize errors commonly associated with the compilation of existing decisions on each ELM architecture. The results show that the proposed model produces an accuracy for the ELM training process of 80.06%. This is coupled with satisfactory value for the ELM testing process of 76.31%.