Articles

NEAR-DUPLICATE REAL-LIFE FACE IMAGE

Main Article Content

Intan Yuniar Purbasari Budi Nugroho

Abstract

NEAR-DUPLICATE REAL-LIFE FACE IMAGE a Intan Yuniar Purbasari, bBudi Nugroho a,bTeknik Informatika, Fakultas Teknologi Industri, UPN “Veteran” Jawa Timur, Indonesia Jl. Raya Rungkut Madya, Gunung Anyar, Surabaya 60294 E-Mail: intan.yuniar@gmail.com Abstrak Content-based Image Retrieval (CBIR) merupakan metode temu kembali citra berdasarkan karakteristik numerik pada citra. Pencarian similaritas yang efisien pada ruang dimensi ultra-high telah diajukan menggunakan two-tier inverted file dan Local Derivative Patterns (LDP) sebagai metode ekstraksi fitur dengan tingkat keakuratan dan kinerja yang tinggi pada data set citra wajah eksperimental. Namun demikian, citra real-life memiliki ukuran dan resolusi yang berbeda serta noise bawaan. Masih belum diketahui apakah LDP dapat menunjukkan hasil yang sama memuaskan jika diberikan data set citra real-life. Penelitian ini merancang dan membangun search engine citra wajah untuk mencari citra nyaris duplikat pada citra real-life menggunakan metode LDP untuk ekstraksi fitur dan two-tier inverted file untuk pengindeks-an multidimensi. Sebuah metode ekpansi state juga diperkenalkan untuk lebih menangkap banyak detil dari histogram citra dengan mempertimbangkan informasi piksel tetangga. Eksperimen ini dilakukan pada 8083 citra wajah real-life dari berbagai ukuran antara 20x20 dan 80x80. Data set berisi kopi duplikat dari citra wajah setelah melalui beberapa proses transformasi. Hasil pencarian mengembalikan 20 citra yang memiliki kemiripan paling tinggi dengan citra query dan memiliki nilai presisi 0.75 atau 75%. Kata kunci: Content-Based Image Retrieval, Local Derivative Pattern, Two-tier Inverted File, Real-life Face Image. Abstract Content-based image retrieval (CBIR) is an image retrieval method based on the analysis of numerical characteristics of the image at the absence of text information. An efficient similarity search in ultra-high dimensional space has been proposed using two-tier inverted file and Local Derivative Patterns (LDP) as feature extraction method with high accuracy and high performance on experimental face image data sets. However, real-life images have different size, resolution and a potential noise. It is unknown whether LDP would show the same satisfactory result given real-life image data sets. This research designed and developed a face search engine to find near-duplicate face in real life images using LDP method to extract image features and two-tier inverted file for multidimensional indexing process. A state expansion method was also introduced to capture more detailed description of image histogram by considering neighbor information. The experiment was performed on 8,083 reallife face images of various sizes between 20x20 to 80x80. The data set contained duplicate copies of face images with some transformation processes. The search result returned top 20 images which had the most similarity with the query images and had an average precision rate of 0.75 or 75%. Keywords: Content-Based Image Retrieval, Local Derivative Pattern, Two-tier Inverted File, Real-life Face Image

Article Details

How to Cite
PURBASARI, Intan Yuniar; NUGROHO, Budi. NEAR-DUPLICATE REAL-LIFE FACE IMAGE. Kursor, [S.l.], v. 7, n. 1, jan. 2013. ISSN 2301-6914. Available at: <https://kursorjournal.org/index.php/kursor/article/view/52>. Date accessed: 12 dec. 2019.
Section
Articles