EarVN1.0
Classification
Face
|...
License: CC BY-NC 3.0

Overview

The EarVN1.0 dataset is constructed by collecting ear images of 164 Asian peoples during 2018. It is among the largest ear datasets publicly to the research community which composed by 28,412 colour images of 98 males and 66 females. Thus, this dataset is different from previous works by providing images of both ears per person under unconstrained conditions. The original facial images have been acquired by unconstrained environment including cameras systems and light condition. Ear images are then cropped from facial images over the large variations of pose, scale and illumination. Several machine learning tasks can be applied such as ear recognition, image classification or clustering, gender recognition, right-ear or left-ear detection and enhanced super resolution.

Data Collection

How Data were acquired

All images are collected and gathered from volunteer's people from 2018 to 2019 in the unconstrained condition such as illumination, occlusion, rotations and mage resolution.

Parameters for data collection

Ear images are cropped from daily and portrait photo semi-automatically.

Description of data collection

This dataset consists of 28,412 images of 164 different peoples.

Value of the Data

  • This is the largest ear images dataset constructed for biometric recognition. Each subject has at least 100 ear images of the left or right side.
  • Automatic ear analysis, including tasks such as ear recognition, person identification, image clustering, imbalanced classification might benefit from this dataset.
  • Some images of this dataset are very low resolution (lower than 25 × 25 pixels) because they are cropped from full facial images. A super-resolution technique could be employed to overcome the inherent resolution limitation.
  • Gender recognition via ear images can be performed and evaluated on this dataset sine we provide 17,571 ear images of male and 10,841 images of female.
  • Right-ear or left-ear detection/recognition can be experimented on this dataset. Moreover, an open problem has been raised if a left-ear image can be matched with a right-ear image of the same person.

Data Annotation

  • Gender recognition/clustering: there are two classes for male and female recognition/clustering tasks. The first 98 folders (from 01 to 98) is belong to male class and the rest (from 99 to 164) is female.
  • Side-ear detection: this is the first open problem for ear recognition by identifying the left-ear or right-ear via image. The potential application of this task can be applied for quick authentication. However, the images are not fully labelled, user can apply the semi-supervised learning for this task.
  • Super-resolution: all ear images of this dataset are on unconstrained low-resolution which has an impact to the performance of biometric systems. We propose to enhance these images to super resolution as an pre-processing step in order to improve the visual clarity and increase the recognition/clustering performance.

Data format

Data format: .jpeg
Type of data: Image in EGB colour space

Citation

@inproceedings{
author={Truong Hoang, Vinh},
year={2020},
title={EarVN1.0},
Mendeley Data, V4, doi: 10.17632/yws3v3mwx3.4
http://dx.doi.org/10.17632/yws3v3mwx3.4
}

License

CC BY-NC 3.0

Data Summary
Type
Image,
Amount
28.412K
Size
--
Provided by
Vinh Truong Hoang
He is an Assistant Professor and Head of Image Processing & Computer Graphics Department, Faculty of Computer Science, Ho Chi Minh City Open University, Vietnam.
Issue
Start Building AI Now