Abstract
Emotional reactions are the best way to express human attitude and thermal imaging mainly used to utilize detection of temperature variations as in detecting spatial and temporal variation in the water status of grapevine. By merging the two facts this paper presents the Discrete Cosine Transform (DCT) with Local Entropy (LE) and Local Standard Deviation (LSD) features as an efficient filters for investigating human emotional state in thermal images. Two well known classifiers, K-Nearest Neighbor (KNN) and Support Vector Machine (SVM) were combined with the earlier features and applied over a database with variant illumination, as well as occlusion by glasses and poses to generate a recognition model of facial expressions in thermal images. KNN based on DCT and LE gives the best accuracy compared with other classifier and features results.
Highlights
Recognition using thermal images has overcome many challenges of the recognition if compared to visible images as illumination [1], [2]; still thermal images faces its own challenges as temperature [3], aging problem and illumination [4], as well as occlusion by glasses and poses which will be tackled in this research
By merging the two facts this paper presents the Discrete Cosine Transform (DCT) with Local Entropy (LE) and Local Standard Deviation (LSD) features as an efficient filters for investigating human emotional state in thermal images
This paper introduces the use of LE and DCT filters as a feature extractors and K-Nearest Neighbor (KNN) as classifier to approach a solution for expression recognition in thermal images
Summary
Recognition using thermal images has overcome many challenges of the recognition if compared to visible images as illumination [1], [2]; still thermal images faces its own challenges as temperature [3], aging problem and illumination [4], as well as occlusion by glasses and poses which will be tackled in this research. Trujillo et al used Local and Global feature extraction methods using interest point detected by Harris detector clustered by K-means with SVM as classifier over IRIS database achieving 76.6% accuracy [5]. Shangfei Wang et al in 2012 introduced temperature difference features and voting strategy with KNN as classifier applied over USTC-NVIE database making 61.62% recognition rate [6]. Deep Boltzmann machine DBM model was used by Shangfei Wang in 2014 for emotional recognition with accuracy rate 62.9% over the USTC-NVIE database [7]. 98.2% recognition rate was achieved by M.H. Abd Latif et al [8] through the use of Gray Level Cooccurrence Matrix (GLCM) as a feature extractor and KNN as a classifier over a new database gathered by the paper team at the International Islamic University in Malaysia
Sign-in/Register to view highlights & summary 
Published Version (
Free)
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call