FIREMAP: A Robust Facial Image Segmentation Framework For Enhanced Emotion Detection in Customer Service Applications

Abstract

The performance and accuracy of the emotion detection system heavily rely on how well face image segmentation and preprocessing are done. This paper proposes FIREMAP, a Facial Image Refinement and Emotion Mapping Architecture, an integrated framework that overcomes the limitations of some of the conventional segmentation methods such as Watershed and K-Means. FIREMAP unifies multi-stage denoising, adaptive illumination correction, and boundary-preserving segmentation to extract fine facial features from input images under non-standard lighting conditions with added noise. Experimental results are provided using ten different facial samples; FIREMAP outperforms all other key metrics and yields the best performance: 96.72% sensitivity, 95.25% specificity, 93.45% Jaccard coefficient, Dice coefficient of 95.83%, and 95.55% accuracy. It has outperformed conventional algorithms by a large margin. These results confirm that FIREMAP outperforms FIRE in terms of robustness, structural consistency, and feature integrity and allow for efficient preprocessing of facial images in real-world customer service environments for emotion recognition applications.