No glasses, hats or scarves were allowed. A total of 16 captures per person were taken in every session, with different poses and lighting conditions, trying to cover all possible variations, including turns in different directions, gestures and lighting changes. In every case only one parameter was modified between two captures. This is one of the main advantages of this database, respect to others. This database is delivered for free exclusively for research purposes.
For each session, three shots were recorded with different but limited orientations of the head. Details about the population and typical problems affecting the quality are given in the referred link. The quality was limited but sufficient to show the ability of 3D face recognition. For privacy reasons, the texture images are not made available. In the period , this database has been downloaded by about researchers. A few papers present recognition results with the database like, of course, papers from the author.
It contains three-dimensional images of facial surfaces. These meshes correspond to 61 different individuals 45 male and 16 female having 9 images for each person. The total of the individuals are Caucasian and their age is between 18 and 40 years old.
Each image is given by a mesh of connected 3D points of the facial surface without texture. The database provides systematic variations with respect to the pose and the facial expression.
Each recording contains a speaking head shot and a rotating head shot. Sets of data taken from this database are available including high quality colour images, 32 KHz bit sound files, video sequences and a 3D model.
LSFM is the largest-scale 3D Morphable Model 3DMM of facial shapes ever constructed, based on a dataset of around 10, distinct facial identities from a huge range of gender, age and ethnicity combinations. A newly created high-resolution 3D dynamic facial expression database are presented, which is made available to the scientific research community.
The 3D facial expressions are captured at a video rate 25 frames per second. For each subject, there are six model sequences showing six prototypic facial expressions anger, disgust, happiness, fear, sadness, and surprise , respectively. Each expression sequence contains about frames. The database contains 3D facial expression sequences captured from subjects, with a total of approximately 60, frame models. Each 3D model of a 3D video sequence has the resolution of approximately 35, vertices.
Well-validated emotion inductions were used to elicit expressions of emotion and paralinguistic communication. Frame-level ground-truth for facial actions was obtained using the Facial Action Coding System.
Facial features were tracked in both 2D and 3D domains using both person-specific and generic approaches. The work promotes the exploration of 3D spatiotemporal features in subtle facial expression, better understanding of the relation between pose and motion dynamics in facial action units, and deeper understanding of naturally occurring facial action.
The database includes 41 participants 23 women, 18 men. An emotion elicitation protocol was designed to elicit emotions of participants effectively. Eight tasks were covered with an interview process and a series of activities to elicit eight emotions.
The database is structured by participants. Each participant is associated with 8 tasks. For each task, there are both 3D and 2D videos. The database is in the size of about 2. Skip to content. Star Branches Tags. Could not load branches. Could not load tags. Latest commit. Git stats 25 commits. Failed to load latest commit information. Nov 15, Nov 12, Nov 8, View code. Several video sequences of varying resolution, conditions and zoom level for each subject. In the analyses below, we exclude those participants who did not complete all cells and note how many participants have been excluded.
It should be noted that for the Strength scale, a large proportion of the data were missing. This was due to many participants rating their internal emotion as being neutral.
As a result, rating the strength of a neutral emotion was inappropriate, and this field was left blank by participants. Overall, the percentage of raters agreeing on the emotion displayed in the images was Descriptive statistics for the different groups can be found in Table 1.
Follow-up ANOVAS revealed that the effect of image category was stronger with black-and-white than with color pictures, but in both cases the positive images were rated as being more positive than the neutral images, which were rated as more positive than the negative images. For more details, please see the supplementary results.
We also observed a main effect of image category, reflecting that the negative and the positive images were rated as being clearer than the neutral images see the supplementary materials ; similar unsurprising main effects were found in other analyses as well, and will be presented only in the supplementary materials. Follow-up analysis showed that females rated the negative and positive images as being more intense than the neutral ones, whereas the males did not vary in their mean intensity ratings across image categories though their means showed numerically the same tendency as those of females.
The same pattern was found with the analysis of clarity, with females rating positive and negative images as being clearer than neutral ones. Numerically, male raters showed the same tendency, but it did not reach significance. Follow-up analyses showed that younger infants elicited internal emotions closer to those intended i.
See Table 3. Follow-up analyses revealed that, for both databases, negative images were rated as being more intense than positive images, which in turn were rated as more intense than neutral images. However, this effect was somewhat more pronounced for the Pearson image set, especially for negative images. A similar pattern was found when clarity was used as the DV, where the negative images in the Pearson image set were rated as being the clearest among the three categories, whereas negative images did not differ in clarity from positive images in the City database.
Follow-up analyses revealed that, for the City database, the positive images were rated as being the most genuine, whereas neutral and negative images did not differ in genuineness.
In contrast, for the Pearson image set, the neutral images were rated as being the least genuine, with no difference between the positive and negative images.
To measure how much the perception of each picture changed over time, the midwifery students were asked to rate a subset of the pictures on two occasions.
An average rating per image for Time 1 and Time 2 was then calculated. To assess how likely participants were to change their minds about the images between Time 1 and Time 2, the number of occasions was counted on which a participant changed her or his mind for the images, and the percentage of changes was calculated relative to the total number of ratings changed and unchanged.
For negative images, participants changed their minds on 1. We compared the ratings across the three groups midwives, neonatal nurses, and general public on all rating scales. A main effect of image category was found for all analyses, and will not be reported further. Follow-up analyses revealed that, for the midwives and the general public, the negative images were rated as more intense than the positive images, which in turn were rated as more intense than the neutral images.
For the neonatal nurses, in contrast, the positive images were rated as the most intense. Follow-up analyses revealed that the midwives and the general public rated the positive and negative images as being clearer than the neutral images, with no difference between the positive and negative images. In contrast, the neonatal nurses rated the positive images as being clearer than the neutral images see Table 4. Follow-up analyses showed that the positive images were rated as being more genuine than the neutral or negative images, with no difference between the latter two categories.
This effect was most pronounced for midwives. Follow-up tests showed that, whereas the images generally elicited the internal emotions expected from the image category i.
The ratings of the strength of the affective response were marginally lower in the general public than in the other groups. The database and norming data can be accessed on request by e-mailing cityinfantfacedatabase gmail. The database contains portrait images, with both black-and-white and color versions available though the color versions have not been fully validated; researchers should take this into account if considering using a mix of the black-and-white and color images.
Color images have not been resized or normalized in terms of their luminosity or hue. In all, 30 of the infants have photographs showing positive, negative, and neutral expressions. In the case of this database, the positive facial expressions are defined as smiling, laughing, or excited; the negative facial expressions are defined as sad, angry, worried, scared, or distressed.
There are a total of 60 positive images, 54 negative images, and 40 neutral images to choose from. Images of 35 girls and 33 boys are included in this database, all from 0 to 12 months of age. Sixty-two of these babies are Caucasian, three are Asian, two Arab, and one Indian. Descriptive statistics, including percentages, can be found in Table 5. For more demographic information about the infants included in this database, please see the online supplemental materials.
This article reports the development and validation of the City Infant Faces database. The results suggest that this database has excellent face validity, with an average agreement rate of The database is comparable to other image sets of infant faces Pearson et al.
Test—retest reliability was also good for all images, although neutral images showed a somewhat higher rate of changes in ratings across time. Additionally, the results showed that neonatal nurses rated the images as being the least genuine and the most positive and as eliciting the internal emotions they expected from the image category, as compared to midwives and the general public.
This suggests that the images should be used with caution in groups of individuals exposed to high levels of extreme infant emotion. Furthermore, it is unclear whether there are consistent differences between the black-and-white and color images with regard to the rating scales.
The majority of the color and black and white images had no significant differences between their ratings. ANOVAs indicated that the black-and-white images were rated as clearer.
This should be taken into account if researchers are considering using a mixture of the color and black and white images in their research. However, there are many inconsistencies in this literature.
Additionally, Geangu, Benga, Stahl, and Striano found that male infants between 1 and 9 months of age cried for longer and more intensely than did female infants. Therefore, it is not clear why this result was found, and future research should look into this.
Eliciting stronger internal emotions and adults seeing the emotion expressed by the infant as more genuine may help the infant to survive. This is because the only way that an infant can survive is through the care of adults, and evoking positive reactions from adults is likely to increase caregiving behavior by the adult Lorenz, ; Luo et al.
A few limitations should be taken into account when using this database. One of the main limitations is that the images were not specifically validated on parents. Although some of the participants who took part may have been parents, this was not measured. Furthermore, because only six males contributed to the ratings for this image set, it is unclear whether this database is valid for use with males.
The results from this database support this showing that females rated the images as more intense and clearer. As a result, caution should be taken if researchers wish to use this database with males. Another limitation of the database is that the majority of the infants are Caucasian. Although significant efforts were made to try and recruit babies of different ethnicities, this was unfortunately not successful.
Furthermore, due to the naturalistic way these images were taken, not all images were taken at the same time. Therefore, although this could be a possible drawback to the database, it is something that could not be overcome when producing such naturalistic images.
The images in this database are arguably more naturalistic than the images from other databases. Furthermore, these images are often taken by a professional photographer under controlled conditions e. The differences in how the images in different databases were produced may explain the findings from this study in terms of the negative images. It could be suggested that all of the negative images in this set were rated as less intense and less clear because of the selection process and production of the images.
For example, the negative images selected by Pearson et al. This could, therefore, be the reason behind the lower ratings for the negative images in this database.
Despite this, the naturalistic nature of these images may be more reflective of infant emotion during parent-infant interaction. This is a clear advantage of the database, since before infants are able to communicate verbally, their facial expressions are one of their main methods of communication. Having naturalistic facial expressions in the database may enable researchers to learn more about the processing of infant emotions that parents are likely to see on a day-to-day basis, rather than extreme emotions that may not be seen as often.
These images may therefore provide researchers with a new way to investigate maternal sensitivity. Thus, despite the limitations of this database, it has many benefits, and therefore can provide a useful tool for researchers to use when researching infant emotion.
Adamson, L. The still face: A history of a shared experimental paradigm. Infancy, 4, — Article Google Scholar. Ainsworth, M. The development of infant-mother attachment. Ricciuti Eds. Google Scholar. Alley, T. Head shape and the perception of cuteness. Developmental Psychology, 17, — Babchuk, W. Sex differences in the recognition of infant facial expressions of emotion: The primary caretaker hypothesis.
Ethology and Sociobiology, 6, 89— Bjorklund, D. FEI Face Database. There are 14 images for each of individuals, a total of images. All images are colourful and taken against a white homogenous background in an upright frontal position with profile rotation of up to about degrees. All faces are mainly represented by students and staff at FEI, between 19 and 40 years old with distinct appearance, hairstyle, and adorns.
The number of male and female subjects are exactly the same and equal to An array of three cameras was placed above several portals natural choke points in terms of pedestrian traffic to capture subjects walking through each portal in a natural way. While a person is walking through a portal, a sequence of face images ie.
Due to the three camera configuration, one of the cameras is likely to capture a face set where a subset of the faces is near-frontal. The dataset consists of 25 subjects 19 male and 6 female in portal 1 and 29 subjects 23 male and 6 female in portal 2. In total, the dataset consists of 54 video sequences and 64, labelled face images. UMB database of 3D occluded faces. The database is available to universities and research centers interested in face detection, face recognition, face synthesis, etc.
The main characteristics of VADANA, which distinguish it from current benchmarks, is the large number of intra-personal pairs order of thousand ; natural variations in pose, expression and illumination; and the rich set of additional meta-data provided along with standard partitions for direct comparison and bench-marking efforts.
It contains , unique images of 40, subjects. Subject ages range from 15 to 80 with a median age of The average number of images per subject is 5 and the average time between consecutive photos is days, with the minimum being 1 day and the maximum being 3, days 9 years, and days.
The standard deviation of days between images is The maximum duration between images for a single subject is 3, days 10 years, and days. The database is licensed for developmental, and commercial uses. It is an abbreviated version, both in terms of images and associated metadata, that is available here. Face images of subjects 70 males and 30 females were captured; for each subject one image was captured at each distance in daytime and nighttime. All the images of individual subjects are frontal faces without glasses and collected in a single sitting.
PhotoFace: Face recognition using photometric stereo. This unique 3D face database is amongst the largest currently available, containing sessions of subjects, captured in two recording periods of approximately six months each. The Photoface device was located in an unsupervised corridor allowing real-world and unconstrained capture. Each session comprises four differently lit colour photographs of the subject, from which surface normal and albedo estimations can be calculated photometric stereo Matlab code implementation included.
This allows for many testing scenarios and data fusion modalities. Eleven facial landmarks have been manually located on each session for alignment purposes. Additionally, the Photoface Query Tool is supplied implemented in Matlab , which allows for subsets of the database to be extracted according to selected metadata e.
The Dataset consists of multimodal facial images of 52 people 14 females, 38 males acquired with a Kinect sensor. The data is captured in two sessions at different intervals of about two weeks. In each session, 9 facial images are collected from each person according to different facial expressions, lighting and occlusion conditions: neutral, smile, open mouth, left profile, right profile, occluded eyes, occluded mouth, side occlusion with a sheet of paper and light on.
An RGB color image, a depth map provided both as a bitmap depth image and a text file containing the original depth levels sensed by Kinect as well as the associated 3D data are provided for all samples.
In addition, the dataset includes 6 manually labeled landmark positions for every face: left eye, right eye, tip of the nose, left side of mouth, right side of mouth and the chin.
Other information, such as gender, year of birth, ethnicity, glasses whether a person wears glasses or not and the time of each session are also available. YouTube Faces Database. The data set contains 3, videos of 1, different people. All the videos were downloaded from YouTube. An average of 2. The shortest clip duration is 48 frames, the longest clip is 6, frames, and the average length of a video clip is In designing our video data set and benchmarks we follow the example of the 'Labeled Faces in the Wild' LFW image collection.
Specifically, our goal is to produce a large scale collection of videos along with labels indicating the identities of a person appearing in each video. In addition, we publish benchmark tests, intended to measure the performance of video pair-matching techniques on these videos. Finally, we provide descriptor encodings for the faces appearing in these videos, using well established descriptor methods.
The dataset consists of subjects, specifically Caucasian females, from YouTube makeup tutorials. Images of the subjects before and after the application of makeup were captured.
There are four shots per subject: two shots before the application of makeup and two shots after the application of makeup. For a few subjects, three shots each before and after the application of makeup were obtained.
The makeup in these face images varies from subtle to heavy. The cosmetic alteration is mainly in the ocular area, where the eyes have been accentuated by diverse eye makeup products. Additional changes are on the quality of the skin due to the application of foundation and change in lip color.
This dataset includes some variations in expression and pose. The illumination condition is reasonably constant over multiple shots of the same subject. In few cases, the hair style before and after makeup changes drastically. We added makeup by using a publicly available tool from Taaz. Three virtual makeovers were created: a application of lipstick only; b application of eye makeup only; and c application of a full makeup consisting of lipstick, foundation, blush and eye makeup.
Hence, the assembled dataset contains four images per subject: one before-makeup shot and three aftermakeup shots. MIW Makeup in the "wild" Dataset. The MIW dataset contains subjects with images per subject. Total number of images is 77 with makeup and 77 without makeup. The images are obtained from the internet and the faces are unconstrained. It currently contains frames of 17 persons, recorded using Kinect for both real-access and spoofing attacks.
Each frame consists of: 1 a depth image x pixels — 1x11 bits ; 2 the corresponding RGB image x pixels — 3x8 bits ; 3 manually annotated eye positions with respect to the RGB image. The data is collected in 3 different sessions for all subjects and for each session 5 videos of frames are captured. The recordings are done under controlled conditions, with frontal-view and neutral expression. In the third session, 3D mask attacks are captured by a single operator attacker.
If you use this database please cite this publication: N. Erdogmus and S. Senthilkumar Face Database Version 1. The Senthilkumar Face Database contains 80 grayscale face images of 5 people all are men , including frontal views of faces with different facial expressions, occlusions and brightness conditions. Each person has 16 different images. The face portion of the image is manually cropped to x pixels and then it is normalized.
Facial images are available in both grayscale and colour images. This database contains video frames of x resolution from 60 video sequences, each of which recorded from a different subject 31 female and 29 male. Each video was collected in a different environment indoor or outdoor resulting arbitrary illumination conditions and background clutter. Furthermore, the subjects were completely free in their movements, leading to arbitrary face scales, arbitrary facial expressions, head pose in yaw, pitch and roll , motion blur, and local or global occlusions.
SiblingsDB Database. The SiblingsDB contains two different datasets depicting images of individuals related by sibling relationships. The first, called HQfaces, contains a set of high quality images depicting individuals 92 pairs of siblings. A subset of 79 pairs contains profile images as well, and 56 of them have also smiling frontal and profile pictures. All the images are annotated with, respectively, the position of 76 landmarks on frontal images and 12 landmarks on profile images. For each individual the information on sex, birth date, age the highest and average age differences between siblings are 30 and 4.
The second DB, called LQfaces, contains contains 98 pairs of siblings individuals found over the Internet, where most of the subjects are celebrities. The position of the 76 frontal facial landmarks are provided as well, but this dataset does not include the age information and human expert ratings were not collected since this dataset is composed mainly of well-known personages and, hence, likely to produce biased ratings.
The Adience image set and benchmark of unfiltered faces for age, gender and subject classification. The dataset consists of 26, images, portraying 2, individuals, classified for 8 age groups, gender and including subject labels identity. It is unique in its construction: The sources of the images included in this set are Flickr albums, assembled by automatic upload from iPhone5 or later smartphone devices, and released by their authors to the general public under the Creative Commons CC license.
This constitutes the largest, fully unconstrained collection of images for age, gender and subject recognition. Large face datasets are important for advancing face recognition research, but they are tedious to build, because a lot of work has to go into cleaning the huge amount of raw data. To facilitate this task, we developed an approach to building face datasets that detects faces in images returned from searches for public figures on the Internet, followed by automatically discarding those not belonging to each queried person.
The FaceScrub dataset was created using this approach, followed by manually checking and cleaning the results. It comprises a total of , face images of celebrities, with about images per person. As such, it is one of the largest public face databases.
Frontalization is the process of synthesizing frontal facing views of faces appearing in single unconstrained photos.
Recent reports have suggested that this process may substantially boost the performance of face recognition systems. This, by transforming the challenging problem of recognizing faces viewed from unconstrained viewpoints to the easier problem of recognizing faces in constrained, forward facing poses.
Authors provide frontalized versions of both the widely used Labeled Faces in the Wild set LFW for face identity verification and the Adience collection for age and gender classification.
These sets, LFW3D and Adience3D are made available along with our implementation of the method used for the frontalization.
Indian Movie Face database IMFDB is a large unconstrained face database consisting of images of Indian actors collected from more than videos. All the images are manually selected and cropped from the video frames resulting in a high degree of variability interms of scale, pose, expression, illumination, age, resolution, occlusion, and makeup. IMFDB is the first face database that provides a detailed annotation of every image in terms of age, pose, gender, expression and type of occlusion that may help other face related applications.
The goal of this project is to mine facial images and other important information for the Wikipedia Living People category. Currently, there are over 0. In addition to these faces, useful meta data are released: the source images, image captions if available , and person name detection results through a named entity detector.
So, mining experiments can also be performed. This is an unique property of this benchmark compared to others. It is a database of 10, natural face photographs of all different individuals, and major celebrities removed. This database was made by randomly sampling Google Images for randomly generated names based on name distributions in the US Census. Because of this methodology, the distribution of the faces matches the demographic distribution of the US e. The database also has a wide range of faces in terms of attractiveness and emotion.
Ovals surround each face to eliminate any background effects. Additionally, for a random set of 2, of the faces, we have demographic information, attribute scores attractiveness, distinctiveness, perceived personality, etc , and memorability scores included with the images, to help researchers create their own stimulus sets. This database contains stereo videos of 27 adult subjects 12 females and 15 males with different ethnicities. The database also includes 66 facial landmark points of each image in the database.
A newly created high-resolution 3D dynamic facial expression database are presented, which is made available to the scientific research community.
The 3D facial expressions are captured at a video rate 25 frames per second. For each subject, there are six model sequences showing six prototypic facial expressions anger, disgust, happiness, fear, sadness, and surprise , respectively. Each expression sequence contains about frames.
The database contains 3D facial expression sequences captured from subjects, with a total of approximately 60, frame models. Each 3D model of a 3D video sequence has the resolution of approximately 35, vertices.
BP4D-Spontanous Database. Well-validated emotion inductions were used to elicit expressions of emotion and paralinguistic communication. Frame-level ground-truth for facial actions was obtained using the Facial Action Coding System. Facial features were tracked in both 2D and 3D domains using both person-specific and generic approaches.
The work promotes the exploration of 3D spatiotemporal features in subtle facial expression, better understanding of the relation between pose and motion dynamics in facial action units, and deeper understanding of naturally occurring facial action.
The database includes 41 participants 23 women, 18 men. An emotion elicitation protocol was designed to elicit emotions of participants effectively. Eight tasks were covered with an interview process and a series of activities to elicit eight emotions. The database is structured by participants. Each participant is associated with 8 tasks. For each task, there are both 3D and 2D videos.
The database is in the size of about 2. The database contains 3D face and hand scans. It was acquired using the structured light technology. According to our knowledge it is the first publicly available database where both sides of a hand were captured within one scan.
Although there is a large amount of research examining the perception of emotional facial expressions, almost all of this research has focused on the perception of adult facial expressions. There are several excellent stimulus sets of adult facial expressions that can be easily obtained and used in scientific research i. However, there is no complete stimulus set of child affective facial expressions, and thus research on the perception of children making affective facial expression is sparse.
In order to fully understand how humans respond to and process affective facial expressions, it is important to have this understanding across a variety of means. The Child Affective Facial Expressions Set CAFE is the first attempt to create a large and representative set of children making a variety of affective facial expressions that can be used for scientific research in this area. The set is made up of photographs of over child models ages making 7 different facial expressions - happy, angry, sad, fearful, surprise, neutral, and disgust.
It is mainly intended to be used for benchmarking of the face identification methods, however it is possible to use this corpus in many related tasks e. Two different partitions of the database are available. The first one contains the cropped faces that were automatically extracted from the photographs using the Viola-Jones algorithm. The face size is thus almost uniform and the images contain just a small portion of background.
The images in the second partition have more background, the face size also significantly differs and the faces are not localized.
The purpose of this set is to evaluate and compare complete face recognition systems where the face detection and extraction is included. Each photograph is annotated with the name of a person. There are facial images for 13 IRTT students. They are of same age factor around 23 to 24 years. The images along with background are captured by canon digital camera of The actual size of cropped faces x and they are further resized to downscale factor 5.
Out of 13, 12 male and one female. Each subject have variety of face expressions, little makeup, scarf, poses and hat also. The database version 1. There are facial images for 10 IRTT girl students all are female with 10 faces per subject with age factor around 23 to 24 years. The colour images along with background are captured with a pixel resolution of x and their faces are cropped to x pixels.
This IRTT student video database contains one video in. Later more videos will be included in this database. The video duration is This video is captured by smart phone. The faces and other features like eyes, lips and nose are extracted from this video separately. Part one is a set of color photographs that include a total of faces in the original format given by our digital cameras, offering a wide range of difference in orientation, pose, environment, illumination, facial expression and race.
Part two contains the same set in a different file format. The third part is a set of corresponding image files that contain human colored skin regions resulting from a manual segmentation procedure. The fourth part of the database has the same regions converted into grayscale.
The database is available on-line for noncommercial use. The database is designed for providing high-quality HD multi-subject banchmarked video inputs for face recognition algorithms.
The database is a useful input for offline as well as online Real-Time Video scenarios. It is harvested from Google image search. The dataset contains annotated cartoon faces of famous personalities of the world with varying profession.
Additionally, we also provide real faces of the public figure to study cross modal retrieval tasks, such as, Photo2Cartoon retrieval. The IIIT-CFW can be used for the study spectrum of problems, such as, face synthesis, heterogeneous face recognition, cross modal retrieval, etc.
Please use this database only for the academic research purpose.
0コメント