NORMAL AND ABNORMAL HUMAN FACE DETECTION BASED ON DCT AND FFT TECHNIQUES- A PROPOSED METHOD

. In today’s world face detection is the most important task. Due to the chromosomes disorder sometimes a human face suffers from different abnormalities. For example, one eye is bigger than the other, cliff face, different chin-length, variation of nose length, length or width of lips are different, etc. For computer vision currently this is a challenging task to detect normal and abnormal face and facial parts from an input image. In this research paper a method is proposed that can detect normal or abnormal faces from a frontal input image. This method used Fast Fourier Transformation (FFT) and Discrete Cosine Transformation of frequency domain and spatial domain analysis to detect those faces.


INTRODUCTION
A face is a key feature to identify a human. During the early stage of embryogenesis development genetic factors playing a key role to develop a face. Due to the chromosomal disorder, facial abnormalities can occur. In this modern era with advanced research technologies, it is possible to identify this facial disorder with help of various methods.
Researchers develop various methods to identify a face. They also develop algorithms to detect different facial parts.
The primary objective of this study is to detect whether a human face is normal or abnormal. To implement this process in this research first face and facial detection system are developed. Then based on the length and the width of the face and facial areas the detection process is developed. This research is a real-world application with a high accuracy rate. The Fast Fourier transformation and Discrete Cosine transformation are used for the precise detection of normal and abnormal humans. The findings of this research suggested that this proposed methodology is useful in this research area and can lead more objective research in the future.

RELATED WORKS
Several researchers implemented several algorithms for face detection with different classification algorithms. For this, they use either a single classification model or a combined model. To detect a face a faked or not researchers used Generative Adversarial Networks (GANs) to automatically detected those fake faces [1]. Some researchers review in their paper about the Down syndrome (DS) disorder of the human, caused by chromosome abnormalities.
They describe face detection, feature extraction, and different classification techniques on it [2]. To detect a face researchers used haar cascade likes and Euclidean distance measurement in their experiment. They show that the recognition rate is 91.1% for the experiment [3]. Researchers proposed a generalized solution to detect visually observable symptoms on faces. For this purpose they used a semi-supervised learning technique with different classification technique algorithms [4]. Other researchers proposed in their research paper an improved technique in Preprints (www.preprints.org) | NOT PEER-REVIEWED | Posted: 26 July 2021 doi:10.20944/preprints202107.0570.v1 face recognition based on Principal Component Analysis and Fast Fourier transformation algorithm. They used the yale database and series of experiments to develop the model [5].

HUMAN FACE STRUCTURE
Each human face has unique features aspects of every individual. In the human body, the face is the key body part for identifying a human. In anatomy human face [6] is divided into three main regions, 1) Upper Face, 2) Middle face and 3) Lower face.

Upper Face
The area started from just below the hairline to the under lower eye-lid, is known as the upper face. This region is divided into two regions.
Forehead. It is the biggest and superior region of the upper face area. This area started from just below the hairline area. It contains two features, skull, and scalp.
Eyes. In the upper face region, the eyes are situated at the semi-circular orbital socket. It is the other superior area of the upper face region. It contains two eye-lids, upper and lower eyelids, and the eyebrows.

Middle Face
The middle face started from the lower eyelid region to just above the upper lip region. This is the central part of a face. It is divided into three parts 1) Nose, 2) Cheek, 3) Ear.
Nose. It is the middle line structure that extends to the face region. In this experiment the nose region divides into two parts, 1) Nose tip and 2) The region from the nose tip to just above the upper lip region.
Cheeks. This region is lateral to the nose. It contains skin and fat pads.
Ears. Humans' ears are used to hear the sound from the environment. This part is the lateral outline of the nose region.

Lower Face
The lower face region is started from the upper lip region to the ching region. This part contains four parts 1)Upper Lip region, 2)Lip region, 3)Chin region, and 4) Jawline region.
Upper Lip Region. This area started from the below nose region to just above the lip region.
Lip Region. This region is started from just below the upper lip region to just above the chin region. This area is also known as the mouth.
Chin Region. This area started from the below lip region to the end of the face region. This is the expanded region of the cheeks.
Jaw Lines. This area is started with the parallel to the nose tip point. The jawline region defines the lower structure of the human face.
All the regions' details described above are shown in figure 1. All the yellow points with red color encircled are the main 8 points of a face and others are used to define those areas more precisely.  In this part, the below steps are followed [7][8][9].
Step 1: Resize the input image.
Step 2: Convert the RGB input color image into grayscale image.
Step 3: To remove any noise from the image apply a filter.
Step 4: To improve the contrast of the filtered image apply the histogram equalization technique.
In this experiment to remove the noise from the grayscale image Gaussian filter is used.

Face Detection
In this experiment to detect a normal or an abnormal human, face detection is the key step. After pre-processing the input image this is the second step. Here to achieve this the Viola-Jones face detection algorithm [10] is used.
This algorithm is very much useful to detect and categorize a human from non-human faces. The accuracy to detect a face using this algorithm is much higher.

Facial Parts Detection
The third step after detecting a face from an image is to detect the different facial parts of that face [11].

Euclidean Distance Calculation
After successfully detecting the face and different facial parts, the fourth step is to calculate the length and width of those regions. To calculate the length and width of those extracted facial features the Euclidean Distance [12] measurement formula shown in equation 3 is used.
Where ( , ) and ( , ) are the two co-ordinates of two points.

Classification Analysis
The final step is to compare the calculated matrices of the training dataset with the test dataset. Based on the threshold value, a binary matrix is created. This matrice is used to determine the normal and abnormal humans from the input facial images. The required algorithms and data flow diagrams are shown in the algorithms and data flow diagram section.

ALGORITHMS AND DATA FLOW DIAGRAM
The algorithms that describe the proposed methodology are given below.
Detection_of_normal_abnormal_humans(Image dataset) [Perform the normal and abnormal human detection]  The preprocessing result of the human face is depicted in figure 5. And detecting different facial parts depicted in figure  6.