Razvoj kombinirane metode za analizo obraznih slik z uporabo sistema sledenja pogleda

Iskra, Andrej (avtor), Gabrijelčič Tomc, Helena (mentor)

Uporaba obraznih slik je v današnjem času zelo razširjena. Z množično uporabo sistemov sledenja pogleda je področje analize gledanja obraznih slik postalo predmet številnih raziskav. Namen naše raziskave je bil razvoj nove kombinirane metode za analizo obraznih slik, s katero bi dodatno preverili rezultate analize obraznih slik, ki so bili pridobljeni po doslej znanih metodah. Prvi korak sta bila izbor in priprava obraznih slik iz ustreznih baz. Zaradi raznovrstnosti zadanih testiranj smo uporabili dve bazi obraznih slik: Minear and Park in Stirling. Najprej smo opravili časovno testiranje s štirimi različnimi časi in tremi različnimi dimenzijami sprednjih obraznih slik. Rezultate opazovanja obraznih slik v povezavi z uspešnostjo prepoznave smo predstavili s kombinirano metodo, ki je sestavljena iz doslej že poznane časovno dolžinske metode in novo zasnovane ploščinske metode. Dodatno smo uporabili metodo analize obraznih delov in metodo merjenja časov odgovorov pri razpoznavi. Vsi rezultati so nam pokazali točko spremembe pri opazovanju obraznih slik štiri sekunde in le pogojno uporabnost sistema sledenja pogleda pri majhnih dimenzijah obraznih slik. Te rezultate smo uporabili za pripravo testov profilnih obraznih slik, pri katerih smo posledično uporabili dve različni dimenziji obraznih slik (srednjo in veliko). Primerjava prepoznave sprednjih in profilnih obraznih slik nam je pokazala občutne razlike med sprednjimi in profilnimi obraznimi slikami le pri napačni prepoznavi. Uporaba kombinirane metode in metode analize obraznih delov pa nam je razkrila drugačen način opazovanja profilnih obraznih slik, saj se je ogled celotnega obraza zgodil v krajšem času kot pri sprednjih obraznih slikah. Naslednje testiranje smo izvedli na slikah obrazov, prikazanih pod različnimi koti, pri čemer so nam rezultati kombinirane metode pokazali različno dojemanje obrazov v odvisnosti od kota prikaza obraza. Na koncu smo kombinirano metodo testirali še pri prepoznavi čustev, pri čemer smo ugotovili veliko skladnost ploščinske metode in metode analize obraznih delov. Ker nam je analiza obraznih delov v večini potrdila rezultate prepoznave čustev, lahko to novo ploščinsko metodo uporabimo tudi pri prepoznavi čustev. Rezultat našega raziskovalnega dela je kombinirana metoda, katere del je novo razvita ploščinska metoda, ki služi kot podpora dosedanjim metodam pri analizah obraznih slik s sistemi sledenja pogleda.

• Jezik: Slovenski jezik
• Ključne besede: sistem sledenja pogleda, obrazne slike, obrazni deli, obrazna čustva, uspešnost prepoznave, fiksacije, sakade
• Vrsta gradiva: Doktorsko delo/naloga
• Organizacija: NTF - Naravoslovnotehniška fakulteta
• Leto izida: 2020
• PID: 20.500.12556/RUL-116791
• Datum objave v RUL: 11.06.2020

Sekundarni jezik

• Jezik: Angleški jezik
• Naslov: Development of combined method for analysis of facial images using eye tracking system

Izvleček

The use of facial images is widespread today. Popularization of the use of eye tracking systems, the field of observing facial images has become the subject of many researchers. The purpose of our research was development a new combined method for the analysis of facial images, which would further verify the results of the analysis of facial images obtained by other methods. The first step was selection and preparation of facial images from chosen facial databases. Due to the variety of tests we have chosen two facial image bases: Minear & Park and Stirling. At the beginning we performed time-spatial analyses with four different times and three different dimensions of the frontal facial images. The results of observing facial images in relation to recognition success were presented by combined method which consist previously known timespatial method and the newly designed area method. Additional results we obtained with facial features method and method of measuring response times in recognition process. All the results showed us the turning point in observing face images at four seconds observation time and indicated potential problems of use of eye tracking systems for small dimensions of facial images. These results were used for the preparation of profile facial image tests, where we used only two different dimensions of facial images (medium and large). Comparison of the recognition of front and profile facial images showed us a significant difference in results between frontal and profile facial images only in false recognition. The use of the combined method and the facial features method revealed a different way of observing profile facial images, where observing of whole face happened in a shorter time than that of the frontal facial images. Next testing was done on faces presented at different view angles, where the results of combined method showed us different perception of those facial images. The combined method was also tested in emotion recognition, where we found a great compliance of the area method and the facial features method. Since the facial features method has mainly confirmed the results of emotion recognition, we can also use the new surface method for emotion recognition. The result of our research work is a combined method, part of which is a newly developed area method, which results supported the methods used so far in the analysis of facial images with eye tracking systems.

• Ključne besede: eye tracking system, facial images, facial features, facial expressions, recognition success, fixations, saccades