M.Sc. Tezi Görüntüleme | |||||||||||||||||||||
|
|
||||||||||||||||||||
Summary: The thesis contains the examination of determining similarity between human faces. Detection of similarity can be defined as listing faces similar to a given face and to be able to achieve it a system should imitate the talent with which humans recognize faces despite a lot of factors. Defining a new way of representing faces taking possible changes into account and classifying a given face using the chosen way are the required operations. To obtain faces to decide how similar they were, firstly a face detection on a given image was performed. In the face detection part of the work, a multilayer neural network was implemented. The network was trained to distinguish faces from non-faces. After training the network, to detect all faces image containing faces was scanned while it was scaled down.In face detection, since possible faces on input image are included on the skin part, skin parts were focused to decrease the image part to be searched for faces. With this aim, to detect skin parts input image was converted into YCbCr color space. The result obtained in the skin detection was processed in examination based on the neural network in the face detection part. Principal Component Analysis is the method preferred in similarity evaluation part of the work. Principal Component Analysis provides an efficient way of comparing faces according to the similarities and differences between them representing faces on a lower dimensional space. Four distance measures, voting based on them and the neural network deisgned were utilized in evaluation of the results obtained in Principal Component Analysis. Additionally, comparing face images by subtraction was performed to comparewith the result obtained using Principal Component Analysis. Listing similar faces in images included in a database designed to be used in similarity evaluation was performed succesfully and most similar faces to a given face which was not incuded in the database could also be listed.Keywords: Face Detection, Artificial Neural Networks, Principal Component Analysis, Image processing. |