An Adaptive Novelty Detection Approach to Low Level Analysis of Images Corrupted by Mixed Noise

Alexander N. Dolia, Martin Lages, Ata Kabán

Abstract

We propose a new adaptive novelty detection based algorithm for the primary local recognition of images corrupted by multiplicative/additive and impulse noise. The purpose of primary local recognition or low level analysis such as segmentation, small object and outlier detection is to provide a representation which could be potentially used e.g. in context based classification or nonlinear denoising techniques. The method is based on the estimation of mixing parameters (priors) of probabilistic mixture models along a small sliding window. A novelty score is defined by the mixing parameters and this is utilized by the procedure for determining the corresponding class of image patch with the aid of a lookup f. Numerical simulations demonstrate that the proposed method is able to improve upon previously employed techniques for the same task. In addition, the computational demand required by the proposed method is clearly inferior to some of the recently applied techniques as expert systems or neural networks.
BibTeX
@inproceedings{dolia2003adaptive,
  author = {Alexander N. Dolia and Martin Lages and Ata Kabán},
  title = {{An Adaptive Novelty Detection Approach to Low Level Analysis of Images Corrupted by Mixed Noise}},
  booktitle = {KES},
  year = {2003},
  doi = {10.1007/978-3-540-45224-9\_78},
}

Loading related content...

Figure