Pengembangan Hybrid Denoising untuk Meningkatkan Kualitas Segmentasi Citra dengan Noise

Meidyani, Biandina (2019) Pengembangan Hybrid Denoising untuk Meningkatkan Kualitas Segmentasi Citra dengan Noise. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 05111750010012-Master_Thesis.pdf Restricted to Repository staff only Download (1MB) | Request a copy

Abstract

Noise pada citra dapat menurunkan kualitas citra dan hilangnya beberapa informasi detail citra. Oleh karena itu, dibutuhkan proses pengurangan noise yang dinamakan denoising citra agar kualitas citra menjadi lebih baik. Namun, masalah utama denoising adalah bagaimana menghilangkan noise dengan tetap mempertahankan tepi. Proses denoising juga bertujuan untuk meningkatkan kualitas segmentasi karena citra yang akan diproses telah memiliki kualitas yang lebih baik.
Penelitian ini bertujuan membangun metode yang efektif agar proses denoising tetap mempertahankan detail dan tepi untuk menghilangkan random-valued noise (gaussian dan speckle noise) maupun fixed-valued noise (salt & pepper noise) dengan melakukan Bilateral Filtering dan Non-local Means Filtering pada koefisien aproksimasi Discrete Wavelet Transform (DWT). Sedangkan untuk koefisien detail akan dilakukan Soft-Thresholding dan diikuti dengan proses Anisotropic Diffusion Filtering. Uji coba dilakukan dengan citra grayscale yang ditambahkan dengan noise. Citra hasil dari metode denoising ini akan dibandingkan dengan metode-metode yang lain seperti: Bilateral Filter, Median Filter, Mean Filter, Wiener Filter dan Gaussian Filter dengan pengukuran MSE dan PSNR. Selanjutnya dilakukan segmentasi Fuzzy C-Means dengan korelasi spasial yang memanfaatkan informasi spasial pada piksel berdasarkan nilai korelasi antar piksel. Iterasi FCM dengan korelasi diinisialisasi dengan hasil pusat cluster dari FCM tanpa modikasi. Analisis kinerja segmentasi menggunakan perhitungan sensitivity, specificity, dan accuracy antara citra groundtruth dengan citra hasil segmentasi. Berdasarkan hasil uji coba dapat disimpulkan bahwa metode hybrid denoising berhasil meningkatkan kualitas segmentasi citra dengan cara menghilangkan noise. Metode ini memiliki rata-rata MSE paling kecil dibandingkan metode lain yaitu sebesar 0,0016 dan rata-rata nilai PSNR paling tinggi dibandingkan metode lainnya yaitu sebesar 28,6758 db.
========================================================== Noise can reduce image quality and lose of some detailed image information. Therefore, a noise reduction process called image denoising is needed so that the image quality becomes better. However, the main problem with denoising is how to eliminate noise while maintaining the edge. The denoising process also aims to improve the quality of segmentation because the image to be processed has better quality. This study aims to build an effective method so that the denoising process maintains detail and edges to eliminate random-valued noise (gaussian and speckle noise) and fixed-valued noise (salt & pepper noise) by performing Bilateral Filtering and Non-local Means Filtering on approximation coefficients of Discrete Wavelet Transform (DWT). While for the detail coefficients, Soft-Thresholding will be carried out and followed by the Anisotropic Diffusion Filtering process. Testing are carried out with grayscale images added with noise. The image results from this denoising method will be compared with other methods such as: Bilateral Filter, Median Filter, Mean Filter, Wiener Filter and Gaussian Filter with MSE and PSNR measurements. Next, Fuzzy C-Means segmentation is carried out with spatial correlation that utilizes spatial information on pixels based on the correlation value between pixels. FCM iterations with correlations are initialized with the results of the cluster center of FCM without modification. Segmentation performance analysis uses calculation of sensitivity, specificity, and accuracy between groundtruth images and segmented images.
Based on the results of the trial it can be concluded that the hybrid denoising method succeeded in improving the quality of image segmentation by eliminating noise. This method has the smallest average MSE compared to other methods which is equal to 0,0016 and the highest average PSNR value compared to other methods which is equal to 28,6758 db.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	denoising citra, noise filtering, segmentasi citra
Subjects:	T Technology > T Technology (General)
Divisions:	Faculty of Information and Communication Technology > Informatics > 55101-(S2) Master Thesis
Depositing User:	Ms Biandina Meidyani
Date Deposited:	23 Jul 2026 03:39
Last Modified:	23 Jul 2026 03:39
URI:	http://repository.its.ac.id/id/eprint/66761

Actions (login required)

View Item