Classification of Facial Expression Recognition using Machine Learning Algorithms

Face is the reflection of brain facial expressions are the discernible consequences of moving a facial muscle. In the perception of human articulations, the facial expression like Sad, Happy, Anger, Disgust, Fear is assumes a fundamental part. The objective of this research focuses on facial expression recognition which related to machine learning and optimization algorithms. Here, we proposed Hybrid Adaptive Kernel based Extreme Learning Machine (HAKELM) algorithm to identify the human facial expression based on certain image processing technique. Thus, the obtained results shown the proposed HAKELM scheme achieved 10% more accuracy, sensitivity and specificity than existing algorithms like Principal Component Analysis (PCA) and Support Vector Machine (SVM).


Introduction
Over recent years, omnipresent computers and digital processing came out to be an extremely vital part of everyday life. It also pushes advances in the creation of agent-controlled interfaces. As an essential element in human life, our feelings and influences also made us communicate and understand our intentions, emotions, and feelings affecting unconsciously our day-to-day activities like thinking, decision making, and interpersonal relations. Therefore, in the current technological era where life is so complicated, the detection of emotions automatically has become a current centre of the AI field, since the importance of impacts in human life and daily functioning are well known. Facial Expression Recognition (FER) System is the invasive and recent successful technology in biometric technology. It can be denoted as verification of individual identifier about the human face. FER is considered as a computer application for automatically identifying or verifying a person in the form of digital image. The Facial Expression Recognition system was introduced during 1978. The importance for evolving FER system was face detection, image normalization, feature extraction, feature analysis and classification (Ming-Hsuan Yang et al. 2002). Every individual has a fairly unique face that can be captured without user co-operation. In face recognition, image processing techniques are used to enhance raw images which are captured from cameras/sensors located on satellites, space probes and aircrafts or pictures from photographs. Over a decade, a FER had gained a significant attention in several applications of pattern recognition, image analysis, commercial identification, marketing tool security systems and biometrics such as fingerprint or eye iris recognition systems (Gosavi&Khot 2013). The facial expression recognition has gained awareness in the growing security concerns to avoid forgery and untreatable activities. In this paper, we proposed the Hybrid Adaptive Kernel based Extreme Learning Machine (HAKELM) algorithm to identify the human facial expression based on certain image processing technique. Thus, the performance results are acquires in terms of accuracy, sensitivity and specificity. Further, the section 2 organises the related works of facial emotional states. Section 3, explains the proposed methodology. Section 4, gives results and discussion. Section 5, concluded the entire work of facial expression recognition.

Related Works
This chapter includes a detailed literature review for the FER system. All the pre-processing, Feature extraction and classification techniques mentioned in the literature are discussed in this section. The machine learning techniques for the Facial Expression Recognition (FER) approach has been proposed (Harihara Santosh Dadi&Gopala Krishna Mohan 2016). In which, pre-processing is employed to 23 diminish the noise. Feature extraction has been performed through the Histogram of Oriented Gradients (HOG) which basically store the edges of the face and the directionality of those edges. Supervised Support Vector Machine is employed to classify the face patterns. The performances' results are tested on two sets of databases such as AT&T database and the YALE B to examine the results which demonstrates that the proposed methods accomplish 90% accuracy.

PROPOSED METHODOLOGY
Our proposed methodology undergoes based machine learning and optimization techniques using the image processing techniques are developed to detect the facial expressions. Figure.1 shows that the overall proposed work of HAKELM-CSO for face detection. The datasets are collected from the available human expression of 256x256 data sets. Pre-processing is done by the wiener filter to reduce the unwanted noise such as salt and pepper that is created during the image acquisition and to improve the image quality. Gesture extraction done by GLCM to extract the particular skin region is facial detection. The extracted gestures of the face skin regions are selected by the modified firefly to reduce the processing time. Finally, the HAKELM technique is utilized to classify and recognize the face detection and image expression which is optimized by 68 CSO for efficient classification. The algorithm steps for HAKELM-CSO are given below. Algorithm steps: Step 1: The facial image of 256x256 is taken from the given each dataset. The given image is read and converted into a gray image.
Step 2: The input images are pre-processed by the wiener filter to remove the noises.
Step 3: The gesture extraction process is done to extract the relevant gestures from the facial expression images by Grey Level matrix.
Step 4: The extracted gesture is selected by Modified Firefly algorithm.
Step 5: The selected images are classified by HAKELM.
Step 6: Chicken Swarm Optimization is used to optimize the kernel parameters evaluated from HAKELM.
Step 7: The resultant facial images are used to classify the facial expressions.
Step 8: Performance measure of Accuracy, Sensitivity and Specificity are compared to estimate the accurate classification of facial expression recognition. .

Results and Discussion
The experimental evaluation of HAKELM with CSO is simulated by the MATLAB software. The main aim of this method is to classify the various facial expressions to identify the patient's expressions. The performance of HAKELM compared with existing algorithms such as SVM (Li Xia 2014) and PCA. Then the trained image are compared with the query image and the facial expressions are classified as normal, happy, sad, surprise, angry and disgust.   Table. 1 shows the comparison analysis of metrics such as Accuracy, Sensitivity and Specificity for the HAKELM and the existing SVM and PCA.

Conclusion
The facial expression recognition analysed using Machine Learning (ML) and optimization algorithms. Hybrid Adaptive Kernel based Extreme Learning Machine (HAKELM) algorithm proposed to identify human facial expression based on certain image processing technique. Thus, the performance results shows the proposed HAKELM scheme achieved a more accuracy of 95.5%, sensitivity of 90.12%, precision of 95.1%, specificity of 96.12%, compared to that of the existing algorithms such as PCA and SVM.