Face Recognition based on CNN 2D-3D Reconstruction using Shape and Texture Vectors Combining

Edy Winarno,Imam Husni Al Amin,Prajanto Wahyu Adi,Sri Hartati

doi:10.11591/ijeei.v8i2.1369

Abstract

This study proposes a face recognition model using a combination of shape and texture vectors that are used to produce new face images on 2D-3D reconstruction images. The reconstruction process to produce 3D face images is carried out using the convolutional neural network (CNN) method on 2D face images. Merging shapes and textures vector is used to produce correlation points on new face images that have similarities to the initial image used. Principal Component Analysis (PCA) is used as a feature extraction method, for the classification method we use the Mahalanobis method. The results of the tests can produce a better recognition rate compared to face recognition testing using 2D images.

Highlights

The human face recognition system is research in the field of computer vision that continues to develop today
The Convolutional Neural Network (CNN) method can produce 3D face images which are processed using a shape and texture combination to produce a correlation of points on a new face image that has similarities to the initial image used
This test shows that the accuracy of facial recognition using Principal Component Analysis (PCA)-Mahalanobis is better than other methods

Summary

INTRODUCTION

The human face recognition system is research in the field of computer vision that continues to develop today. Several studies on the development of facial recognition have been carried out by several previous researchers such as [1,2,3,4,5] Some of these face recognition studies include many using facial recognition processing using 2-dimensional (2D) imagery in the process of image data acquisition, preprocessing, feature extraction and classification. From several previous face recognition studies, there are still many methods and algorithms that have not been studied, the use of reconstruction algorithms from 2D images to 3D forms that are used as databases in face recognition This 2D to 3D image reconstruction method is expected to contribute robustly in face detection and recognition so that it has high accuracy and fast facial recognition computing. The results of the process of combining vector shapes and textures from 3D face images are processed using a database for the face recognition process

RESEARCH METHOD

Shape and Texture Combining

Mahalanobis Distance

RESULTS AND DISCUSSION

Findings

CONCLUSION