CMED: A Child Micro-Expression Dataset

Nikin Matharaarachchi, Muhammad Fermi Pasha, Sonya Coleman, and Kah Peng Wong · 2025

Read CMED: A Child Micro-Expression Dataset by Nikin Matharaarachchi, Muhammad Fermi Pasha, Sonya Coleman, and Kah Peng Wong (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-21690.txt.

Introduction

Micro-expressions are brief flashes of feeling that cross a face too fast to hide, and therapists read them for clues. Nikin Matharaarachchi, Muhammad Fermi Pasha, Sonya Coleman, and Kah Peng Wong note that research on these flashes covers adults only. Children's expressions differ, and no data set of child micro-expressions existed, partly because children are hard to record in a controlled way.

The team builds the first such data set, called CMED, from spontaneous videos captured during online video calls. With real child examples in hand, they compare child and adult micro-expressions and point out key differences. They also set a first benchmark for spotting and naming these expressions automatically, trying three approaches that mix hand-built features and learned models. The data set opens a new line of work for emotion-aware tools that support child therapy.

Catalog record

CREATORNikin Matharaarachchi, Muhammad Fermi Pasha, Sonya Coleman, and Kah Peng Wong
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE35 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2503-21690

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

CMED: A Child Micro-Expression Dataset Nikin Matharaarachchi, Muhammad Fermi Pasha, Sonya Coleman, and Kah Peng Wong Abstract Micro-expressions are short bursts of emotion that are difficult to hide. Their detection in children is an important cue to assist psychotherapists in conducting better therapy. However, existing research on the detection of micro-expressions has focused on adults, whose expressions differ in their characteristics from those of children. The lack of research is a direct consequence of the lack of a child-based micro-expressions dataset as it is much more challenging to capture children’s facial expressions due to the lack of predictability and controllability. This study compiles a dataset of spontaneous child micro-expression videos, the first of its kind, to the best of the authors knowledge. The dataset is captured in the wild using video conferencing software. This dataset enables us to then explore key features and differences between adult and child micro-expressions. This study also establishes a baseline for the automated spotting and recognition of micro-expressions in children using three approaches comprising of hand-created and learning-based approaches. I Introduction Micro-expressions (ME) are brief, spontaneous expressions that occur when someone suppresses an emotion. They are very difficult to control or hide. Micro-expressions were first discovered by Haggard and Isaacs [ 1 ] . They were greatly expanded by psychologist Paul Ekman [ 2 ] , who identified six basic emotions that could be expressed through micro-expressions: anger, disgust, fear, happiness, sadness, and surprise. Studies have shown that, from childhood, we learn that specific emotional responses conveyed through facial expressions are appropriate in a given situation. Thus, humans have been trained to hide their genuine emotions in situations due to social expectations or to deceive others deliberately [ 3 ] . Owing to the short-lasting and involuntary nature of micro-expressions, even when people try to hide their genuine emotions by forcibly altering their macro-expressions, such as through the use of smiles, micro-expressions still present [ 4 ] . The ability to hide true emotions by forcing fake facial expressions has been studied widely. It has shown that the ability to judge the authenticity of emotions is vital in many fields such as education [ 5 ] , forensic investigations [ 6 ] , health [ 7 ] [ 8 ] , immigration, customs and security checks [ 9 ] [ 10 ] , court rulings [ 11 ] , business [ 12 ] , politics [ 13 ] , robotics and human-computer interactions [ 14 ] . Child emotions differ fundamentally from adults due to psychological and physiological differences as shown in previous psychological studies [ 15 ] . Mandy et al. [ 16 ] have studied differences between adults and children in expressing surprise and find that children have more pronounced brow and mouth movements due to developmental differences and the lack of emotion regulation. In addition children are prone to rapid emotional changes. Furthermore, collecting samples from children comes with additional challenges due to the lack of predictability and controllability. In the context of micro-expression detection in children, it benefits psychotherapists in performing technology-enhanced therapy. This can be especially beneficial for children who struggle to verbalize their emotions, such as those with autism spectrum disorder (ASD), anxiety, or trauma. In addition, automated micro-expression detection systems integrated into teletherapy platforms could allow therapists to monitor emotions even in remote settings. This is particularly valuable for children in rural areas or during situations where in-person sessions may not be feasible. However, to the best of our knowledge, there has been no attempt to create a dataset for child micro-expressions, and all samples in the primary micro-expression datasets are from adult subjects; thus, work on child micro-expression recognition is non-existent yet high beneficial. While micro-expressions in children may not have been widely studied, the number of studies on micro-expression recognition in adults has been increasing rapidly. Most of the early research has used hand-designed models for feature extraction [ 17 ] . They can be classified into geometric-based features such as Main Directional Mean Optical Flow (MDMO) [ 18 ] and appearance-based features such as Local Binary Pattern- Three Orthogonal Planes (LBP-TOP) [ 19 ] , and 3D Histogram of Oriented Gradients (3D-HOG) [ 20 ] . However, with the advances in deep learning, recent studies focus on using various learning-based approaches to recognize micro-expressions. The primary approach initially was to use a CNN and provide inputs such as a single frame [ 21 ] such as the apex frame [ 22 ] or a sequence of interpolated frames. Studies have also been performed using pre-processed features such as optical flow [ 23 ] , optical strain [ 24 ] and synthetic images [ 25 ] as inputs. More recently, graph-based approaches have been gaining popularity using its variants such as DGCNNs [ 26 ] and T-GCNs [ 27 ] . II Existing Micro-Expression Datasets Micro-expression recognition is the classification of a given set of frames (or single frame) consisting of a micro-expression from one of the emotion classes. While no micro-expression datasets exists for children, the primary datasets used for adult micro-expression recognition tasks are CASME [ 28 ] , CASME II [ 29 ] , CAS(ME) 2 [ 30 ] , SMIC [ 31 ] and SAMM [ 32 ] . The number of samples present and the number of classes in each of the datasets can be found in Table I . In summary: TABLE I: Details of the existing micro-expression datasets. Dataset Subjects Samples Classes FPS CASME 35 195 8 60 CASME II 35 247 5 200 CAS(ME) 2 22 53 (and 250 macro) 4 30 SMIC 16 164 3 100 SAMM 32 159 7 200 CAS(ME) 3 100 1109 (and 3190 macro) 7 30 4DME 41 267 5 60 • The Chinese Academy of Sciences Micro-expression (CASME) database contains 195 micro-expressions filmed under 60 fps. The study included 195 samples from 35 subjects. The samples are divided into amusement, sadness, disgust, surprise, contempt, fear, repression, and tense. • The Chinese Academy of Sciences Micro-expression II (CASMEII) database used higher temporal (200fps) and spatial resolution (about 280x340 pixels on facial area). It consisted of 247 micro-expressions from happiness, disgust, surprise, repression, and other. • The Chinese Academy of Sciences Macro-Expressions and Micro-Expressions (CAS(ME)²) contains both macro-expressions and micro-expressions. It contains 53 expression samples that were recorded at 30 fps, and the resolution was set to 640 \times 480 pixels. The samples were classed as Positive, Negative, Surprise and Other. • The \text{CAS(ME)}^{3} dataset consists of 1,109 samples that were classed as happiness, disgust, surprise, fear, anger, sadness, and other. The samples were collected in 30fps and a resolution of 1280 \times 720 pixels • The Spontaneous Micro-Facial Movement Dataset (SAMM) consists of 159 micro-expressions, captured at a resolution of 2040 × 1088 pixels and 200 frames per second. A key feature of SAMM is its tailored emotional stimuli for each participant, inducing natural responses in a controlled lab setting. The samples were classed as anger, disgust, fear, happiness, sadness, surprise, and contempt. • The Spontaneous Micro-expression Database (SMIC) were collected from 16 participants at 100 fps and were classified into positive, negative and surprise. The dataset also contains a variant with 71

Read the complete text