Loading...

Skin detection using contourlet texture analysis

Fotouhi, M ; Sharif University of Technology

877 Viewed
  1. Type of Document: Article
  2. DOI: 10.1109/CSICC.2009.5349608
  3. Abstract:
  4. A combined texture- and color-based skin detection is proposed in this paper. Nonsubsampled contourlet transform is used to represent texture of the whole image. Local neighbor contourlet coefficients of a pixel are used as feature vectors to classify each pixel. Dimensionality reduction is addressed through principal component analysis (PCA) to remedy the curse of dimensionality in the training phase. Before texture classification, the pixel is tested to determine whether it is skin-colored. Therefore, the classifier is learned to discriminate skin and non-skin texture for skin colored regions. A multi-layer perceptron is then trained using the feature vectors in the PCA reduced space. The Markov property of images is addressed in post-processing to join separate neighbor skin detected regions. Comparison of the proposed method with other state-of-the-art methods shows a lower false positive rate with a little decrease in true positive rate. ©2009 IEEE
  5. Keywords:
  6. Contourlet coefficients ; Contourlets ; Curse of dimensionality ; Dimensionality reduction ; False positive rates ; Feature vectors ; Markov property ; Multi layer perceptron ; Nonsubsampled contourlet ; Post processing ; Reduced space ; Skin detection ; Skin textures ; Skin-colored regions ; State-of-the-art methods ; Texture analysis ; Texture classification ; Training phase ; True positive rates ; Pixels ; Textures ; Vector spaces ; Principal component analysis
  7. Source: 2009 14th International CSI Computer Conference, CSICC 2009, 20 October 2009 through 21 October 2009, Tehran ; 2009 , Pages 367-372 ; 9781424442621 (ISBN)
  8. URL: http://ieeexplore.ieee.org/document/5349608/?reload=true