Pengenalan Citra Porno Berbasis Kandungan Informasi Citra (Image Content)

I. Gede Pasek Suta Wijaya • I. B. K. Widiartha
Journal article Jurnal Teknik Elektro Universitas Kristen Petra • September 2005 Indonesia

Abstract

Pornographic image recognition is a matching process between an image signature of pornographic training image and an image signature of query image. The aims of this research were to build pornographic image recognition system based on image content that can classify an image that is porno or not and to know how the performance of this method is. Image content that was used in this research was color content and image signature. The image content was taken by color histogram and by extracting an image using wavelet transform, next choosing a little wavelet transforms coefficients that have the biggest magnitude value and moment was used to sharpness the image information content. The tests were carried out in Daubechies 8 wavelets type, 25 heterogeneous pornographic images for training and 500 query images that consist 125 heterogeneous pornographic images and 375 non-pornographic images. Screening success rates were determined by using number images that can be screening per number images on database in percentage, and screening time was determined by the time that needed for an image to classify as pornographic image. The results show that this method give average of screening success rate about 67.02% (it means can classify 84 images as pornographic from 125 pornographic images in database) and can classify about 36 images as pornographic from 375 non-pornographic images in database. This system also needs a little time for screening process that is about 0.29 second every image and screening time is linear to size of database. This method is good enough as pornographic image screening but need some improvement for increasing the performance.

Metrics

  • 368 views
  • 95 downloads

Journal

Jurnal Teknik Elektro Universitas Kristen Petra

Jurnal Teknik Elektro Universitas Kristen Petra is published biannually, in May and September, by... see more