Accurate Face Tagging with Efficient Regularizer and Distance Metric Algorithm
Deepak Sudhakar Dharrao, Nilesh Juke · 2018 International Conference On Advances in Communication and Computing Technology (ICACCT) · 2018
In computer vision detection and recognition of accurate image with great efficiency in real time is a milestone for researchers. The aim of proposed system is to recognise an accurate image from a large set of database and refer a correct name to each face detected in each frame. In many situation every image frame contains a more than one face. To assigning a true name to every face in each frame, firstly it is important to detect a faces from an image is foremost task. In this system for face detection we have used the Viola Jones algorithm. To accomplish our objective of correct face naming normalization of image and distance vector plays an important role. Two isolated affinity matrices are designed for the same. L1 norm based regularizer is used to create a first matrix and ambiguously supervised structural metric learning algorithm is used to create another matrix. After creating two matrices they are fused to create fused affinity matrix which gives proper name to each face in fame. Mahalanobis distance of the data is used to form a matrix with ASML algorithm. While in training sample if large no. of unrelated features is there an L1 norm regularizer is more effective. Finally fusion of these two affinity matrices is used to assign a correct name of each face at proper position of in the frame.