Return to Databases

Tunisian Horses DataBase of Regim Lab’2015 (THoDBRL’2015)

Tunisian Horses DataBase of Regim Lab’2015 (THoDBRL’2015)

 

1. Introduction

Animal recognition is an active research topic in recent years. Horse’s recognition is an important task in the world and  in  order  to  promote  horse’s  recognition  research,  the  Tunisian  Research  Groups  in  Intelligent  Machines  of University of Sfax (REGIM of Sfax) will provide the THoDBRL’2015 database freely of charge to mainly horses’ face recognition researchers and to increase total of researches done to enhance animal recognition. This Database is used in [1]. The main objective of the construction of this database is to identify horses keeping its natural behavior without any direct ontact between the animal and the camera.

2. THoDBRL’2015 Database:

THoDBRL’2015 database is a multiview horses’ faces database, which is created in March 2015. Images of 47 horses’ faces were captured from four equestrian centers in Sfax (a south town in Tunisia) during 3 days in daylight. The size of the database is 4.67Go and contains 2820 images.

3. Camera setup:

The capture is done by the video camera using a digital still camera of 10.1 Mega Pixels and at a resolution of 640 × 480 pixels. The camera is hand-held and positioned in front of the photographer eyes at a distance of about 10 cm from his face. These digital images were taken at a distance of about 1 m from horses when horses in the barn under natural conditions.

4. Capture Conditions:

Horses in motion: The animal is not sane and it is impossible to fix his head and keep its stability. It may change its place and position. These changes of place spoil the distance of 1 meter between the camera and the animal. In order to fix utmost the head of the animal to obtain adequate images, we chosen to take captures for horses when they are in barns. The movements of the head and body of the horse are decreasing, but no longer disappear. We tried to capture different views, to change the background image and to keep the distance of 1 meter between the camera and the animal, but because of the motion of the animal, this distance may vary a one animal to another and between the same animal pictures. So, facial capture of this database present a few changes distance from the camera.

Accessories horses: We choose to capture horses without bridle that we consider as noise for horses identification. There is only the horse number three in our database with a red bridle. It refused to be closer to him to strip the bridle. The horses’ faces data was captured from 3 views; view profile, view right and view left of the horse.

Natural conditions: There is the luminance degradation in the horse’s face according to the position of the animal head and the position of the sun. Also, there is the presence of the shadow like the shadow of the walls or leaves. The background in each captured video varies from one horse to other.

5. Data collection procedure:

Since we cannot fix the horse to keep it stable during the capture, we choose to capture video for each horse of about 50 seconds to obtain adequate poses. Based on human observation, the best photo is selected with different poses in terms of position of the head and in terms of background and luminance. Selected images are not 100% front and not 100% profile; the view can be inclined a little to the right or to the left. Consequently, the facial images of our database have almost the same size and the same resolution and the amount of difference is not very large.

6. Database organization:

Our THoFDRL’2015 database contains 470 frontal face images, 470 left profile images and 470 right profile images. In fact, we tacked 10 frontal face images, 10 left profile images and 10 right profile images per horse.

In the THoDBRL’2015 database there are four datasets: Dataset F (frontal view dataset), Dataset ProfileR (profile right view dataset), Dataset ProfileL (Left profile view dataset) and Videos. The Dataset F contains two different folders; The first folder named “CutImages” and the second folder named “Images”.

6.1. Dataset F:

  • Images:

This folder “Images” contains frames selected on frontal view (10 frontal images for each horse).

The format of the image filename in this dataset “Image” is ‘xx-mm-tt.jpg’, where

–  xx: direction, xx=01; the id of the frontal view.

–  mm: horse id, from 01 to 47.

–  tt: number image, from 01 to 10

  • CutImages:

We still provide 2 types of horses faces cropped from selected frames of the dataset “Images”. Images of the second type are used in [1].

The format of the image filename in Dataset “CutImages” is ‘xx-kk-mm-tt.jpg’, where

–  xx: direction, xx=01; the id of the frontal view.

–  kk: number of the type; there are two types 01 and 02.

–  mm: horse id, from 01 to 47.

–  tt: number image, from 01 to 10.

6.2. Dataset ProfileR and Dataset ProfileL:

Dataset ProfileR contains frames selected on right profile view of horses (10 right profile images for each horse).

Dataset ProfileL contains frames selected on right profile view of horses (10 left profile images for each horse).

The format of the image filename in Dataset ProfileR and Dataset ProfileL is ‘xx-mm-tt.jpg’, where

–  xx: direction, xx=02 is  the id of the right profile view and xx=03 is the id of the left profile view.

–  mm: horse id, from 01 to 47.

–  tt: number image, from 01 to 10.

6.3. Videos

This dataset is compsed by the captured videos of the 47 horses. These videos are used to selected horses images on the three views. The size of this dataset is 4.63 Go.

Bibliography

[1]: I. Jarraya, W. Ouarda, M. A. Alimi, ‘A Preliminary Investigation on Horses Recognition Using Facial Texture Features’, IEEE SMC, 2015.

 

To download Tunisian Horses DataBase of Regim Lab’2015 (THoDBRL’2015), please download the pdf release agreement by this <link> , fill-in and scan it, then Attach it in the form https://goo.gl/forms/aSUOfWWhAd