Extracting Relative Face Naming Similarity on Low Superiority Image Using Mil

Labelling faces in the images is tremendously challenging due to huge variation in image appearance of each character and the weakness, ambiguity of available annotation. Some of the preprocessing steps to be carried out before performing face naming. Even though successfully performing preprocessing step such as face detector and entity name detector, face naming is still a challenging task. So, the supervised approach requires sufficient labelled training data in order to achieve high accuracy. The supervised learning method such as Multiple Instance Learning algorithm would solve the ambiguity of the labelling process by making weaker assumptions about the labelling information. Based on Multiple Instance Learning algorithm, the two discriminative probabilistic learning method, “Quasi-Positive Bags” and “Extended Diverse Density” is used to develop an automatic training scheme. This algorithm will be applicable for large image databases and images does not have complete data labels.