(For USM Staff/Student Only)

EngLib USM > Ω School of Electrical & Electronic Engineering >

Image enhancement for face images in frequency domain/Tan Pin Yang

Image enhancement for face images in frequency domain_Tan Pin Yang_E3_2013_NI
Dalam pemprosesan imej digital, teknik peningkatan imej sering digunakan sebagai alat pra-pemprosesan untuk memperbaiki imej bagi meningkatkan ketepatan sistem tersebut. Teknik penapisan homomorphik adalah salah satu kaedah peningkatan imej yang digunakan dalam domain frekuensi. Walaupun pelbagai jenis penapis laluan tinggi telah dicadang untuk teknik penapisan homomorphik, penapis-penapis konvensional ini tidak direka secara khusus untuk pemprosesan imej-imej wajah. Demi meningkatkan prestasi teknik penapisan homomorphik pada imej-imej wajah, penapis baru perlu dicipta berdasarkan lokasi komponen pencahayaan dalam spektrum Fourier. Memandangkan komponen pencahayaan dan pantulan dalam domain frekuensi tidak diasingkan dengan jelas, lokasi komponen pencahayaan pada imej-imej wajah perlu dikenal pasti dahulu sebelum mencipta penapis baru. Justeru, satu peralatan telah dicipta dalam kajian ini untuk menangkap imej wajah yang mengalami pencahayaan yang lemah serta sumbangan komponen pantulan yang rendah agar lokasi sebenar komponen pencahayaan boleh dikenal pasti dengan tepat sekali. Dua kaedah telah dicadangkan untuk mengenal pasti lokasi komponen pencahayaan dalam kawasan frekuensi rendah dan kawasan menegak di spektrum Fourier. Setelah itu, kaedah purata pencahayaan dan tiga jenis penapis telah dicadangkan berdasarkan lokasi komponen pencahayaan yang diperolehi. Kelebihan dan kelemahan bagi kaedah dan penapis-penapis yang dicadangkan telah dinilai. Kaedah purata pencahayaan telah ditolak kerana tidak berupaya untuk memperbaiki imej-imej wajah yang mengalami pencahayaan yang buruk. Demi mengesahkan keupayaan dan prestasi penapis-penapis yang dicadangkan, analisis kualitatif dan kuantitatif telah digunakan. Analisis kualitatif membuktikan bahawa penapis terima jalur frekuensi lebih baik daripada kaedah biasa dan penapispenapis yang lain kerana ia berupaya untuk mengurangkan hingar. Penapis terima jalur frekuensi masih kekal sebagai pilihan terbaik walaupun sesetengah masukan imej wajah tidak dapat dikesan oleh perisian pengiktirafan muka. Dalam analisis kuantitatif, kualiti imej wajah yang diperbaiki tidak terbukti lebih baik daripada masukan imej wajah disebabkan beberapa faktor luaran seperti hingar dan tahap kecerahan imej muka. Namun, prestasi penapis terima jalur frekuensi lebih baik daripada penapis konvensional dan penapis lain kerana ia dapat mengekalkan kekayaan maklumat tanpa menghilangkan kontras dan ketajaman imej wajah. Justeru, penemuan-penemuan membukti bahawa penapis terima jalur frekuensi boleh menggantikan penapis lulus tingi dalam teknik penapisan homomorphik. ___________________________________________________________________________________ In digital image processing, image enhancement methods are often used as preprocessing tools to enhance images for better accuracy of a system. Homomorphic filtering technique is one of the image enhancement methods implemented in frequency domain. Despite different derivatives of high pass filters are proposed for homomorphic filtering technique, these conventional filters are not specifically designed for face images. In order to improve the performance of homomorphic filtering technique on face images, a customized filter have to be designed based on the location of illumination components in Fourier spectrum. Since the illumination and reflectance components are not distinctly separated in the frequency domain, the location for illumination components of face images must be identified before designing the customized filter. Thus, in this research, a casing is designed to capture poorly illuminated face images with low contribution of reflectance components so that the exact location of illumination components can be identified accurately. Two methods are proposed to identify the locations of illumination components at the low frequency and vertical centre regions in the Fourier spectrum. Then, illumination averaging method and three filter designs are proposed based on the location of illumination components obtained. The advantages and drawbacks of the proposed method and filters are evaluated during the implementation. Illumination averaging method is rejected due to its limited capability in enhancing most of the poor illuminated face images. In order to verify the capability and performance of the proposed filters, qualitative and quantitative analyses are applied. From qualitative analysis, it is proven that band pass filter outperformed the conventional and other proposed filters due to its noise reduction capability. It also remains as the best option although some of the input face images are hardly detectable by the face recognition software. In the quantitative analysis, the quality of enhanced face images are not statistically proven to be better than that of the input face images due to some external factors such as noise and brightness level of the face image. However, by evaluating the enhanced face images, band pass filter performed better than the conventional and proposed filters as it is able to maintain the richness of information without losing the contrast and sharpness of the face images. Hence, the findings prove that band pass filter can replace the conventional high pass filter in homomorphic filtering technique.
Contributor(s):
Tan Pin Yang - Author
Primary Item Type:
Final Year Project
Language:
English
Subject Keywords:
digital image ; homomorphic ; filter
First presented to the public:
6/1/2013
Original Publication Date:
2/14/2020
Previously Published By:
Universiti Sains Malaysia
Place Of Publication:
School of Electrical & Electronic Engineering
Citation:
Extents:
Number of Pages - 138
License Grantor / Date Granted:
  / ( View License )
Date Deposited
2020-02-14 16:03:45.883
Submitter:
Nor Hayati Ismail

All Versions

Thumbnail Name Version Created Date
Image enhancement for face images in frequency domain/Tan Pin Yang1 2020-02-14 16:03:45.883