| dc.description.abstract | In recentdecades, research in Face Recognition (FR) has grownveryrapidly due to the broaddemand on face recognition systems. Researchers in this area attempt to tackle the difficultiesthat can affect the accuracy of FR. Extracting relevant information from face images was the first research topic in thisarea;manyapproaches have been proposed for thispurposesuch as Principle Component Analysis (PCA), Linear Discriminant Analysis (LDA), DiscreteCosineTransform (DCT), and SparseRepresentation Coding (SRC). In addition to that, researchersattempt to deal withsomeotherfactorsthat can negatively affect the recognition accuracyincluding: facial expressions, illumination (lighting conditions), pose variations, occlusion and distance. Toomanyresearchpapers have been introduced to solve theseproblemsresulting in very high recognition rate. However, each single researchfocusesonly on one factor discardingotherfactors. Unfortunately, face recognition systemsdon’t have the ability of recognisingmost of the abovefactors, especially, illumination.
In thispaper: First , wewill show the shortcomings of the currentresearch in dealingwith the abovementionedfactors. Secondly, wewillclearlydefinedifferent information that can beobtainedfrom face images, namelydiscriminate information and similarity/shared information. Then, wewillillustrate how this information can beusedeffectively to increase Face Recognition rate. Thirdly, based on the previouswork on face recognition research, wewill show that the embedded information such as Local Binary Pattern (LBP) can beimprovedefficiently for FR. Suchimprovement can beachieved by effective extraction of embedded information and good estimation of the value of featuresobtainedfromthis information. Finally, we propose adding a new distinct component to the structure of current face recognition systems, i.e. Facial State Recognition (FSR), including pose, illumination, occlusion, and distance, therefore, convenient FR method can beusedbased on FSR. | en_US |
