Implementasi Metode Hybrid Saliency-Svm Untuk Pemilihan Data Training Secara Otomatis Dalam Segmentasi Citra

Yuliandari, Aisha (2015) Implementasi Metode Hybrid Saliency-Svm Untuk Pemilihan Data Training Secara Otomatis Dalam Segmentasi Citra. Undergraduate thesis, Institut Technology Sepuluh Nopember.

[img]
Preview
Text
5111100150-Undergraduate Thesis.pdf - Published Version

Download (4MB) | Preview

Abstract

Segmentasi citra adalah salah satu tahapan penting dalam visi komputer dan pengolahan citra, yang nantinya akan digunakan untuk temu kembali citra, pengenalan objek dan klasifikasi data. Segmentasi citra bisa dilihat sebagai masalah klasifikasi, yaitu dengan menandai masing-masing piksel menurut ciri-ciri tertentu. Support Vector Machine (SVM) adalah metode klasifikasi yang termasuk dalam supervised learning. Supervised learning merupakan metode yang membutuhkan training dan testing. Training sample yang digunakan pada proses training tidak selalu ada pada beberapa kasus, terutama kasus segmentasi citra. Penelitian ini mengimplementasikan metode berbasis SVM yaitu Saliency-SVM untuk pemilihan data training secara otomatis dalam segmentasi citra. Metode ini menghasilkan data training menggunakan visual saliency berbasis SVM di mana terdapat tahapan pra-segmentasi dan pembentukan trimap berdasarkan informasi saliency dari visual saliency detection, kuantisasi ruang warna HSV, analisis histogram dan local homogeneity threshold. Data training yang dihasilkan adalah piksel yang termasuk positif(objek) dan negatif(background). Tahap sebelum dilakukan viii segmentasi dengan SVM adalah ekstraksi fitur untuk menghasilkan vektor input pada SVM. Segmentasi objek pada citra dilakukan dengan SVM berdasarkan SVM Trained Model. Hasil uji coba dari Saliency-SVM untuk segmentasi citra ini memiliki nilai rata-rata akurasi mencapai 94,84% dibandingkan dengan citra ground truth. ======================================================================================================== Image segmentation is one important step in computer vision and image processing, which will be used form image retrieval, object recognition, and data classification. Image segmentation can be seen as classification problem, namely by marking each pixel with certain characteristics. Support Vector Machine (SVM) is a classification method that is included in supervised learning. Supervised learning is a method that requires training and testing. Training samples which used in learning process does not always exist in some cases, especially in image segmentation cases. This research implements SVM bases method that is Saliency-SVM for choose training data automatically in image segmentation. This method generates training data using visual saliency based SVM, where there are pre-segmentation and trimap generation based on saliency information of visual saliency detection, HSV color space quantization , histogram analysis and local homogeneity threshold. Training data is generated pixels which including positive (object) and negative (background).Step before use segmentation with SVM is feature extraction to generate x input vectors in SVM. Object segmentation in image is done by SVM based SVM Trained Model. The test result of saliency-SVM for image segmentation has an average value accuracy reached 94.84% by comparison premises ground truth image.

Item Type: Thesis (Undergraduate)
Additional Information: RSIf 006.4 Yul i
Uncontrolled Keywords: Segmentasi citra, SVM, visual saliency detection, pemilihan data training.
Subjects: T Technology > TA Engineering (General). Civil engineering (General) > TA1637 Image processing--Digital techniques
Divisions: Faculty of Information and Communication Technology > Informatics > 55201-(S1) Undergraduate Thesis
Depositing User: Mr. Tondo Indra Nyata
Date Deposited: 14 May 2019 02:17
Last Modified: 14 May 2019 02:17
URI: http://repository.its.ac.id/id/eprint/63015

Actions (login required)

View Item View Item