Fan-beam Projection-based Feature Extraction for Facial Expression Recognition

A.Sherly Alphonse

doi:10.54963/dtra.v1i1.31

Abstract

This paper presents a novel method of feature extraction using Fan beam projection-based data. The Fanbeam projection covers the image completely and hence gathers all the important information. Even though the image quality is distorted, this type of feature extraction method helps to gather all the important information as there is a huge volume of projection data. Also, the use of multiple detectors speeds up the entire process. All the projections of the image together form a sinogram image which is unique for each facial expression image. Hence, the sinogram image is divided into grids and the histogram formation results in a feature vector for each image. The classification of these feature vectors using Radial Basis Function-based Extreme learning Machine (RBF-ELM) results in high classification accuracy for all the datasets.

Highlights

Feature extraction is the process through which the essential and non-redundant data are extracted from a raw image that increases the between-class distance and decreases the within-class distance
This may leave out some of the needed information. When these feature vectors are used for the classification of facial expressions, there is a reduction of accuracy
This paper proposes a novel method of extracting features called fan-beam projection-based feature extraction

Summary

Introduction

Feature extraction is the process through which the essential and non-redundant data are extracted from a raw image that increases the between-class distance and decreases the within-class distance. LDP and MRDTP filter the image using Kirsch masks and the prominent edges are considered for the calculation. This may leave out some of the needed information. This paper proposes a novel method of extracting features called fan-beam projection-based feature extraction. This method considers the projection data taken from an image at different angles. Sensors used and a number of angles used for the rotation of the fan-beams This projection data captures unique information from the raw image. It is converted to feature vector by observing the micro-patterns in the projection data. These feature vectors can group the images from the same class very effectively

Materials and Methods

Experimental Results and Discussion

Conclusions