Date on Master's Thesis/Doctoral Dissertation


Document Type

Doctoral Dissertation

Degree Name

Ph. D.


Electrical and Computer Engineering

Committee Chair

Farag, Aly A.

Committee Co-Chair (if applicable)

Alphenaar, Bruce

Committee Member

Sahoo, Prasanna

Committee Member

Inanc, Tamer

Committee Member

Yampolskiy, Roman Vladimirovich


Computer vision; Human face recognition (Computer science)


From a computer vision point of view, the image is a scene consisting of objects of interest and a background represented by everything else in the image. The relations and interactions among these objects are the key factors for scene understanding. In this dissertation, a mathematical model is designed for the detection of partially occluded faces captured in unconstrained real life conditions. The proposed model novelty comes from explicitly considering certain objects that are common to occlude faces and embedding them in the face model. This enables the detection of faces in difficult settings and provides more information to subsequent analysis in addition to the bounding box of the face. In the proposed Selective Part Models (SPM), the face is modelled as a collection of parts that can be selected from the visible regular facial parts and some of the occluding objects which commonly interact with faces such as sunglasses, caps, hands, shoulders, and other faces. With the face detection being the first step in the face recognition pipeline, the proposed model does not only detect partially occluded faces efficiently but it also suggests the occluded parts to be excluded from the subsequent recognition step. The model was tested on several recent face detection databases and benchmarks and achieved state of the art performance. In addition, detailed analysis for the performance with respect to different types of occlusion were provided. Moreover, a new database was collected for evaluating face detectors focusing on the partial occlusion problem. This dissertation highlights the importance of explicitly handling the partial occlusion problem in face detection and shows its efficiency in enhancing both the face detection performance and the subsequent recognition performance of partially occluded faces. The broader impact of the proposed detector exceeds the common security applications by using it for human robot interaction. The humanoid robot Nao is used to help in teaching children with autism and the proposed detector is used to achieve natural interaction between the robot and the children by detecting their faces which can be used for recognition or more interestingly for adaptive interaction by analyzing their expressions.