Contribution Of Different Types Of Facial Information Towards Face Recognition In Different Image Resolution

Face is a part of our social life. It provides us with many kinds of information,such as people’s age, gender, and race, etc. A closer observation can even tell us theirmood, intention, and attentiveness. Through face, we recognize each other,communicate with each other, and in this way, live a better social life.Faces, though tiny, attract the academic interest from many scholars，leading tolots of discoveries, such as face inversion effect, left hand advantage effect and genderdifference in face recognition, as well as the particularity of face recognition. So far,cognitive mechanism of face recognition has been discussed, and several facecognitive models have been put forward. In addition, contribution of different types offacial information towards face recognition has been explored with different resultsachieved.Based on the conclusions of previous research, this paper classifies facialinformation into three categories: internal features, external features andconfigurational information. Internal features include eyebrows, eyes, nose, mouthand other facial organs; external features consist of facial outline, ears and hair, etc;configurational information comprises the spatial structure of internal features,namely the relative position of eyebrows, eyes, nose and mouth.In order to explore types of facial information which people rely on in real life,this study uses color face photos as experimental material, and simulates facerecognition in reality by blurring the faces.There are two experiments in this study. Experiment one investigates facerecognition of subjects in a situation where images of familiar faces, in differentresolutions and with different facial information are presented. The experiment wasdesigned to be a two-factorial within-subject study, in which the facial informationvariable has four levels: internal features without configurational information, external features, internal features with configurational information, and the wholeface. The blur variable also has four levels: Gaussian blur of4.5,3,1.5and clearimage. The materials of experiment one are all photos of famous individuals, subjectsthen are asked to identify the faces based on the face information presented, andreport the name of persons in the photos. The design and processing of materials inexperiment two is similar to that of experiment one, the difference being that thematerials of experiment two are photos of strangers to the subjects. In experiment two,the subjects are first instructed to learn some new faces. And then, in the recognitionstage, they are asked judge whether the presented image is a new face or not.The experimental results show that, in the case of familiar faces, internal featureswith configurational information are more useful at low resolution, whereas, in thecase of unfamiliar faces, external features are more useful.Further data analysis shows that, internal features without configurationalinformation are sensitive to image resolution and the performance of subjects declinesrapidly when recognizing increasingly blurred faces. In contrast, external features areless sensitive to image resolution and the performance does not change significantlywith image resolution. Besides, in cases when full faces are presented, theperformance of face recognition does not change along with the image resolutioneither.