Multi-Featured and Fuzzy-Filtered Machine Learning Model for Face Expression Classification

摘要

Face expression is the appearance-based descriptive information, which is used for recognizing the emotion, behavior, and intention of an individual. In this paper, the content and textural information are utilized for the identification of facial-expression. For exploring the content features, the Gabor filter is applied to normalized face-image. The face and Gabor-face images are divided into smaller blocks. For each block, content, structure, and texture-sensitive quantitative features are extracted. This stage transformed the image into the wider content-adaptive quantitative featureset. The fuzzy-based composite filter is applied to this larger featureset for the identification of the most relevant featureset. SVM classifier with different kernels is applied to this reduced-featureset for accurate recognition of expression. The experimental validation is conducted on JAFFE and CK+ datasets. Analytical observations are collected using accuracy, sensitivity, specificity, FNR, and FPR parameters. The experimentation results show that the proposed model outperformed the state-or-art methods and achieved a significant recognition rate.