Visual Feature Selection Using Saliency Map

DOI Web Site
  • MOCHIZUKI Ryuugo
    Department of Human Intelligence Systems, Kyushu Institute of Technology
  • ISHII Kazuo
    Department of Human Intelligence Systems, Kyushu Institute of Technology

Bibliographic Information

Other Title
  • サリエンシーマップを用いた画像特徴点の選択

Description

<p>Many saliency maps in visual image researches are proposed as human attention models. In general saliency maps, three image feature extractions, color contrast, edge gradient and intensity are utilized and DoG (Difference of Gaussian) is applied for each image feature to have a saliency map. We are focusing on selecting stable feature extraction to have stable saliency map. Although various sizes of objects exist in an image, stability of feature extraction is very important in various scales of images. If the DoG filter scale is improper, feature extraction does not work stably and also important feature can be missed after feature selection. In order to solve the problems, we propose to calculate the scale values which maximize differences between two different scale DoG images’ scales so as to important objects are extracted.</p>

Journal

Keywords

Details 詳細情報について

Report a problem

Back to top