Please use this identifier to cite or link to this item:
https://ptsldigital.ukm.my/jspui/handle/123456789/476366Full metadata record
| DC Field | Value | Language |
|---|---|---|
| dc.contributor.advisor | Khairuddin Omar, Prof. Dr. | |
| dc.contributor.author | Mansaf M Saleh Solman (P49072) | |
| dc.date.accessioned | 2023-10-06T09:17:08Z | - |
| dc.date.available | 2023-10-06T09:17:08Z | - |
| dc.date.issued | 2011-07-11 | |
| dc.identifier.other | ukmvital:84193 | |
| dc.identifier.uri | https://ptsldigital.ukm.my/jspui/handle/123456789/476366 | - |
| dc.description | Face detection is a challenging computer vision problem. Given a still image or an image sequence, the goal of face detection is to locate all regions that contain a face regardless of any 3D transformation and lighting condition. There are two main solutions for this problem: feature-based and image-based approaches. In this thesis, two different image-based and learning oriented solutions are compared, to observe the learning dynamics and face detection performances. In the first approach, named skin color based face detector, used Modeling the distribution of skin color to identify areas most likely to be regions of the skin, so as to identify potential areas of skin by equalizing the probability of likelihood, the problem space is assumed to be linearly separable and, a linear threshold function is offered for the solution which is supported by a sparse feature mapping architecture. For the second approach, a neural network in the form of a multilayer perceptron with back propagation solution is used which assumes to represent any function using arbitrary decision surfaces by utilizing nonlinear activation functions. Observations in the comparative experiments show that the methods show closer performances for the classification in the face and non-face space. The first proposed method in this thesis has been enhanced by combination of these two approaches and has achieved a result of 97.3% which can be considered as a high result compared to second methods.,Mengesan wajah adalah masalah visi komputer yang mencabar. Diberikan imej kaku atau jujukan imej, matlamat pengesanan wajah ialah untuk mengesan semua rantau yang mengandungi wajah seseorang tanpa mengira mana-mana transformasi 3D dan keadaan pencahayaan. Terdapat dua penyelesaian untuk masalah ini: iaitu pendekatan berasaskan ciri dan berasaskan imej. Dalam tesis ini, dua pendekatan berasaskan imej dan orientasi pembelajaran dibandingkan, bagi melihat prestasi kedinamikan pembelajaran dan pengecaman wajah. Pendekatan pertama, disebut sebagai pengesan wajah berasaskan warna kulit, menggunakan model taburan warna kulit untuk mengenal pasti kawasan-kawasan berkemungkinan kawasan-kawasan kulit, supaya dapat mengenal pasti kawasan kulit berpotensi dengan menyamakan kebarangkalian kemungkinan, ruang masalah diandaikan sebagai boleh pisah secara linear dan, fungsi ambang linear diketengahkan sebagai penyelesaian yang disokong oleh seni bina pemetaan ciri jarang. Bagi pendekatan kedua, rangkaian neural dalam bentuk perceptron multiaras dengan penyelesaian perambat-balik digunakan yang dianggapkan mewakili sebarang fungsi menggunakan permukaan kata putus arbitrari dengan memanfaatkan fungsi pengaktifan tak linear. Pemerhatian terhadap perbandingan uji kaji menunjukkan prestasi kaedah-kaedah ini tidak jauh beza untuk mengelaskan wajah dan bukan-wajah. Kaedah cadangan yang pertama dalam tesis ini telah ditambah oleh gabungan dua pendekatan-pendekatan ini dan telah mencapai 97.3% yang boleh ditimbangkan sebagai pencapaian tertinggi berbanding dengan kaedah yang kedua,Master/Sarjana | |
| dc.language.iso | eng | |
| dc.publisher | UKM, Bangi | |
| dc.relation | Faculty of Information Science and Technology / Fakulti Teknologi dan Sains Maklumat | |
| dc.rights | UKM | |
| dc.subject | Biometric identification | |
| dc.subject | Neural networks (Computer science) | |
| dc.title | An enhanced face detection method using skin color and back-propagation neural network | |
| dc.type | theses | |
| dc.format.pages | 74 | |
| dc.identifier.callno | TA1650.S665 2011 tesis | |
| dc.identifier.barcode | 001965 | |
| Appears in Collections: | Faculty of Information Science and Technology / Fakulti Teknologi dan Sains Maklumat | |
Files in This Item:
| File | Description | Size | Format | |
|---|---|---|---|---|
| ukmvital_84193+SOURCE1+SOURCE1.0.PDF Restricted Access | 1.47 MB | Adobe PDF |  View/Open |
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.