Scene classification using pLSA with visterm spatial location

  • Authors:

  • Sunki Kim;Daijin Kim

  • Affiliations:

  • Pohang University of Science and Technology, Pohang, Kyungbuk, South Korea;Pohang University of Science and Technology, Pohang, Kyungbuk, South Korea

  • Venue:
  • IMCE '09 Proceedings of the 1st international workshop on Interactive multimedia for consumer electronics
  • Year:
  • 2009

Quantified Score

Hi-index 0.00

Visualization

Abstract

This paper proposes a novel scene classification method that represents a scene as a latent aspect distribution using the probabilistic latent semantic analysis (pLSA) with visterm spatial location and determines the class label of the input image's latent aspect distributions using the support vector machine (SVM). The proposed scene classification method consists of two stages; training and test stage. The training stage performs the following tasks; (1) we extract local feature descriptors of training scenes using SURF or SIFT. (2) we cluster local feature descriptor using k-means clustering method to obtain visterms. (3) we make spatial weight map of each visterm. (4) we make visterm-image co-occurrence table using spatial weight map. (5) we learn latent aspect probability P(zjI) and visterm probability P(vjz) from using EM method, and we obtain support vector of aspect features P(zjI). The test stage performs the similar procedure with the training stage to obtain aspect features P(zjIin) and determines the class label of input image using k-NN or SVM. Experimental results show that the classification rate of the proposed scene classification method is about 86.19% on the 1,239 scene images.