Development of Multi-Label Image Classification System by Subset Family of Images and Local Features

2014 Fiscal Year Final Research Report

• Project/Area Number: 24500228
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Research Field: Perception information processing/Intelligent robotics
• Research Institution: Kushiro National College of Technology
• Principal Investigator: TENMOTO Hiroshi 釧路工業高等専門学校, 情報工学科, 准教授 (80321371)
• Project Period (FY): 2012-04-01 – 2015-03-31
• Keywords: パターン認識 / 画像分類 / 機械学習 / 計算知能 / 進化計算 / 国際情報交換

Outline of Final Research Achievements

In image recognition system, we often generate histograms based on local image patterns and classify the images by using the histograms as feature vectors. However, natural images usually contain multiple objects in each image, so the generated histogram are of the mixture of the histograms come from such multiple objects. Therefore, we proposed a new randomized algorithm that enumerates and covers the image set by the subset family of images and local features, which may give multi-labels to each image. Through computer simulations, we could confirm that classification results along with the users' sensibility and important visual words can be obtained by both supervised and unsupervised classification methods.

Free Research Field

パターン認識