Loading...
Seismic Image Denoising by Thresholding Neural Network in Curvelet Domain
Haghighatgoo, Leila | 2012
533
Viewed
- Type of Document: M.Sc. Thesis
- Language: English
- Document No: 43535 (55)
- University: Sharif University of Technology, International Campus, Kish Island
- Department: Science and Engineering
- Advisor(s): Haj Sadeghi, Khosro
- Abstract:
- Predicting the location of oil and gas recourses is the first challenge in the petroleum industry. One of the most popular and acceptable ways which can guide an explorer to the position of the resources is the seismic survey. By this kind of survey, geologists can observe inside of solid matter by using the ultrasound waves. The process works by sending sound waves to the surface and measuring the length it takes to be reflected from rocks underneath, then with recording these echoes by arrays of sensors, they can obtain a seismic image which has too noise, including ghosting, multiples (multiples are the waves that has been reflected more than once between the energy source and the receivers.) In this thesis, we have transferred seismic images to the new domain known as curvelet, in which the signal representation is sparser than other transforms. Therefore, denoising the image in this domain is easier than other domains. Also, by working in curvelet domain, we are able to reconstruct our denoised image with higher quality specially in edges that play important role in seismic image interpretation.
By considering the inherent of the seismic image noise; which is unknown and untraceable, one of the best candidates for noise removal in this condition can be an Artificial Neural Network. This kind of filtering can pursue nonlinear behavior effectively. Also signal representation in curvelet domain is sparse. Therefore, thresholding can represent a good result in image denoising. Considering these two facts, we have used the special kind of Artificial Neural Network known as Thresholding Neural v Network introduced by Zhang in which the inputs of the net are image curvelet coefficients and by applying appreciate threshold value to the input, in output of the net we will have just those coefficients that do not include any noises. . By applying this approach we can achieve a better visual result and at least 7.25 dB improvements in Peak Signal to Noise Ratio, PSNR, than other customary methods and therefore, denoise the image more efficiently - Keywords:
- Neural Network ; Curvelet Transform ; Image Denoising ; Thresholding ; Thresholding Neural Network (TNN) ; Seismic Imaging
- محتواي پايان نامه
- view