Classification Methods for CT-Scanned Carcass Midsections - A Study of Noise Stability |
|
| Abstract | Computed tomography (CT) has successfully been applied in medical environments for decades. In recent years CT has also made its entry to the industrial environments, including the slaughter houses. In this paper we investigate classification methods for an online CT system, in order to assist in the segmentation of the outer fat layer in the mid- section of CT-scanned pig carcasses. Prior information about the carcass composition can potentially be applied for a fully automated solution, in order to optimize the slaughter line. The methods comprises Markov Random Field and contextual Bayesian classification, and are adapted to use neighbourhood information in 2D and 3D. Artificial Poisson noise is added to the provided dataset to determine how well each of the meth- ods handles noise. Good noise handling will allow for scanning at lower energies. The investigated methods did not perform better than the refer- ence model in terms of classification, but the MRF segmentation showed promising results in a case with extreme simulated noise. |
| Keywords | Fat segmentation, CT-scan, Food processing, Segmentation of noisy data |
| Type | Conference paper [With referee] |
| Conference | Scandinavian Workshop on Imaging Food Quality, in conjunction with Scandinavian Conference of Image Analysis, Sweden, May 2011 |
| Editors | Proceedings of Scandinavian Workshop on Imaging Food Quality |
| Year | 2011 Month May |
| Electronic version(s) | [pdf] |
| BibTeX data | [bibtex] |
| IMM Group(s) | Image Analysis & Computer Graphics |