|M.Sc Student||Goldman Arnon|
|Subject||Anomaly Subspace Detection Based on a Multi-Scale Markov|
Random Field Model
|Department||Department of Electrical Engineering||Supervisor||Professor Israel Cohen|
Automatic target detection in natural scene, is a challenging problem due to the large variability of background clutter and object appearance. Anomaly detection methods are employed for target detection when no a priori information about the targets is available. The statistics of the background data is often described by random field models such as the Gaussian Markov random field (GMRF).
In many natural clutter images, scene elements often appear to have several periodical patterns, of various period lengths. In such cases, conventional random field models may not sufficiently describe the background clutter. Furthermore, in real detection problems, some partial information may be available in terms of a subspace in which the target signals lie.
In this work, we introduce a multi-scale Gaussian Markov random field (GMRF) model and a corresponding anomaly subspace detection algorithm. Using the Karhunen-Loeve transform (KLT) we generate from a multi-scale representation of the image, independent layers. We assume that these independent layers can be modeled as GMRFs with different sets of parameters. The detection is subsequently carried out by using a modification of the matched subspace detector (MSD). The MSD was originally developed for signal detection in subspace interference and white Gaussian noise. Here we formulate a MSD for subspace signal detection in clutter, which follows the multi-scale GMRF model. A quantitative performance analysis shows the advantages of the proposed method. The proposed model and algorithm are applied to synthetic as well as real images, demonstrating the robustness and flexibility of the algorithm in adverse environments.