Noise Filtering dengan Soft Weighted Median Filter untuk Meningkatkan Kualitas Segmentasi Citra

Manek, Siprianus Septian (2018) Noise Filtering dengan Soft Weighted Median Filter untuk Meningkatkan Kualitas Segmentasi Citra. Masters thesis, Institut Teknologi Sepuluh Nopember.

[thumbnail of 5116201039-TESIS.pdf]
Preview
Text
5116201039-TESIS.pdf - Accepted Version

Download (4MB) | Preview

Abstract

Salah satu faktor penghambat pada proses pengolahan citra adalah noise. Noise pada citra dibedakan menjadi dua jenis yaitu fixed-valued noise (salt & pepper noise) dan random-valued noise (gaussian, poisson, speckle, dan locarvar noise). Penelitian-penelitian sebelumnya yang terkait dengan noise filtering lebih fokus pada fixed-valued noise, sedangkan untuk random-valued noise masih jarang dilakukan.
Penelitian ini mengusulkan metode Soft Weighted Median Filter (SWMF) untuk menghilangkan fixed-valued maupun random-valued noise. Untuk setiap piksel pada citra, langkah pertama yang dilakukan adalah menentukan window 3×3 untuk mencari piksel center dan piksel tetangganya. Kemudian semua nilai piksel pada window tersebut diurutkan dan dibagi menjadi tiga bagian, jika nilai piksel center berada pada bagian kedua, maka dianggap sebagai piksel bebas noise, sedangkan jika nilai piksel center berada pada bagian pertama atau bagian ketiga, maka dianggap sebagai piksel ber-noise. Langkah terakhir pada proses ini adalah mengganti nilai piksel ber-noise dengan nilai rata-rata median tertimbang dari semua piksel dalam window, sedangkan piksel bukan noise dibiarkan tidak berubah. Nilai piksel baru dari proses ini digunakan kembali untuk perhitungan berikutnya.
Citra hasil dari metode SWMF dibandingkan dengan metode-metode yang lain seperti; Median Filter, Mean Filter, Wiener Filter dan Gaussian Filter lewat pengukuran Mean Squared Error (MSE) dan Peak Signal to Noise Ratio (PSNR). Proses segmentasi citra dilakukan pada citra hasil noise filtering, terdiri dari 2 proses yaitu deteksi area (Top-Hat Transform) dan deteksi garis (Sobel Edge Detection). Analisis kinerja pada tahap ini menggunakan perhitungan sensitivity, specificity, dan accuracy antara citra groundtruth dengan citra hasil segmentasi.
Berdasarkan hasil uji coba, dapat disimpulkan bahwa metode Soft Weighted Median Filter berhasil meningkatkan kualitas segmentasi citra dengan cara menghilangkan menghilangkan fixed-valued maupun random-valued noise, metode ini memiliki rata-rata nilai PSNR paling tinggi dibandingkan metode lainnya yaitu sebesar 29,21 db.
==========================================================================================================
One of the inhibiting factors in image digital processing is noise. Noise in the image is divided into two types: fixed-valued noise (salt & pepper noise) and random-valued noise (gaussian, poisson, speckle, and locarvar noise). Previous studies of noise filtering focus on fixed-valued noise, while random-valued noise is rarely done.
This research proposes a Soft Weighted Median Filter (SWMF) method to remove fixed-valued dan random-valued noise. For each pixel in the image, the first step is determine the 3×3 window to search the center pixel and neighboring pixels. Then all pixel values in the window are sorted and divided into three parts, if the the center pixel value in the part two, it is considered as noise-free pixel, whereas if the center pixel value in part one or part three, it is considered as noise pixel. The final step in this process is replace the noise-pixel value with the average of median weighted value of all pixels in the window, while the noise-free pixel are left unchanged. The new pixel value from this process is reused for the next pixel calculation.
The result images of the SWMF method are compared with other methods such as: Median Filter, Mean Filter, Wiener Filter and Gaussian Filter with the measurement of Mean Squared Error (MSE) and Peak Signal to Noise Ratio (PSNR). Image segmentation process is done on the image of noise filtering result. There are two image segmentation process, firstly, area detection using Top-Hat Transform, and secondly, line detection using Sobel Edge Detection. Performance analysis at this stage using the calculation of sensitivity, specificity, and accuracy between groundtruth images with the image of the results of segmentation.
Based on the experiment results, Soft Weighted Median Filter method succeeded to improve the quality of image segmentation by eliminating fixed-valued and random-valued noise. This method has the highest average PSNR value compared to other methods of 29.21 db.

Item Type: Thesis (Masters)
Additional Information: RTIf 006.42 Man n-1 3100018074506
Uncontrolled Keywords: median filter, noise filtering, segmentasi citra
Subjects: T Technology > TA Engineering (General). Civil engineering (General) > TA1637 Image processing--Digital techniques
T Technology > TR Photography
Divisions: Faculty of Information and Communication Technology > Informatics > 55101-(S2) Master Thesis
Depositing User: Siprianus Septian Manek
Date Deposited: 06 Mar 2018 07:45
Last Modified: 08 Jul 2020 12:10
URI: http://repository.its.ac.id/id/eprint/49731

Actions (login required)

View Item View Item