Πλοήγηση ανά Συγγραφέας "Maridaki, Anna"

Τώρα δείχνει 1 - 1 of 1
  • Μικρογραφία εικόνας
    Τεκμήριο
    Categorical assessment of depression based on low level features.
    (Τ.Ε.Ι. Κρήτης, Σχολή Τεχνολογικών Εφαρμογών (Σ.Τ.Εφ), ΠΜΣ Πληροφορική και Πολυμέσα, 2018-06-01) Maridaki, Anna; Μαριδάκη, Άννα
    Mental illness is a disease which usually causes behavior disturbances. Nowadays, mental disorders are commonplace, affecting a large number of people. There are more than 200 forms of mental illness. Depression, bipolar disorder, dysthymia and anxiety disorder are some of the most common mood disorders that are inherently related to emotions. According to the European Statistical System [1], 7.1% of the European citizens reported having chronic depression. Whereas, the World Health Organization ranks major depressive disorder as the 4th leading cause of disability worldwide. Major depressive disorder (MDD), also known as clinical depression, is a mood disorder involving bad mood, low self-esteem and loss of interest in normal pleasurable activities. Depression affects the way of thinking, feeling and acting in daily duties. Major depressive disorder has observable behavioral symptoms. Actually, some behavioral cues are closely associated with depression, such as body language, speech, head and face movement. In most cases, facial expressions indicate individual’s depressed feelings. Technology has the potential to assess those cues for depression diagnosis. There are numerous research groups focused on automated depression detection based on audio and visual signals analysis. Automatic recognition of human emotions have a lot of applications in Human Computer Interaction and Affective Computing field. However until now the performance of such applications is still not satisfying. The aim of this work is to develop a framework for the assessment of major depressive disorder, as a supportive application for the clinical care of patients. In so doing we utilize the dataset of the depression recognition sub-challenge of the Audiovisual Emotion Recognition challenge (AVEC), as it is the only publically available dataset consisted by video recordings that are annotated with a depression index. Algorithmic methods were developed, using MATLAB, for the detection of depression using low level image-based features. In the proposed framework, two different motion representation methods were tested, namely: a) Motion History Image (MHI), and b) Gabor Motion History Image (GMHI). Appearance-based descriptors employed were the employed, namely Local Binary Pattern (LBP), Local Phase Quantization (LPQ), and Histogram of Oriented Gradients (HOG). Additionally, some statistical features were utilized. Subsequently, machine learning algorithms were selected in order to define the specificity of the proposed framework. The selected classification algorithms are k-Nearest Neighbors (kNN), Random Forest, Support Vector Machine (SVM) and Naïve Bayes. For the MHI approach the best performance 81.93% achieved by appearance-based descriptor HOG, and with combination of HIST-MEAN-STD using SVM classifier for 100 selected features. The execution time for those results are 1.46 seconds and 0.15 seconds respectively. As for GMHI approach the maximum F1 score is 81.93% achieved by the combination of statistical features HIST-MEAN-STD for 10 selected features and the execution time is 0.12 seconds.