A Robust Design of Real-Time Resilient Smile Recognition System

Mr. T. KATAIAH 1
Assistant Professor, Departmentof AIDSAnnamacharya Institute ofTechnology and Sciences,
Tirupati – 517520, A.P.kata09573@gmail.com
K .YAMINI 4
UG Scholar, Department ofAIDSAnnamacharya Institute ofTechnology and Sciences,
Tirupati – 517520, A.P.konayamini5@gmail.com
K. RAVI SHANKAR 2
UG Scholar, Department of AIDSAnnamacharya Institute ofTechnology and Sciences,
Tirupati – 517520, A.P.shankerr270@gmail.com
M. VARSHA 5
UG Scholar, Department of AIDSAnnamacharya Institute ofTechnology and Sciences,
Tirupati – 517520, A.P.varsha1232004@gmail.com

S. NIKHIL 3
UG Scholar, Department of AIDSAnnamacharya Institute ofTechnology and Sciences,Tirupati – 517520, A.P.sankarapunikhilkumar@gmail.com

Abstract— Facial expressions are an integral part of human-computer interaction. Smiles are used to convey  emotions such as happiness, relaxation and comfort in daily-life situations. So, building a system that can detect smiles accurately is useful for many real-life applications. In this work, we propose a real-time smile recognition system by using both deep learning and classical machine learning models. The proposed systemfirst uses a Haar Cascade-based face detector and then image preprocessing such as grayscale conversion, resizing and normalization. Classical machine learning models extract features using the Histograms of Oriented Gradients (HOG) method, whereas deep learning models learn features directly from the data. Several models such as CNN model, MobileNetV2 model, and SVM,Random Forest and KNN algorithms are used, so that we can compare the performance and understand the prosand cons. A unified-prediction-setup is designed, which selects among the different models to predict whether a  person is smiling or not. Additionally, a confidence score is added to the proposed output. Effectiveness of thedifferent models is assessed using standard performance metrics such as accuracy, precision, recall, F1-score and confusion matrix. From the obtained results, we advise that deep learning models have higher accuracy, while traditional models are faster and require fewer computational resources. Finally, the system is implemented for real-time use using webcam input, which is useful for smart cameras, monitoring systems and interaction platforms. Overall, the proposed system seeks to find a balance between accuracy and efficiency, while having a good performance in real-life scenarios.