Quantification of computed tomography pork carcass images

Estimation of lean meat percentage from computed tomography (CT) images of scanned carcasses is the basis for grading meat quality. Due to noise, artifacts and partial volume effects, the automatic classification of tissues and its posterior quantification is difficult. In this paper we present a new processing pipeline that integrates a partial volume model to classify the CT pork carcasses in three tissues: fat, lean, and bone. The approach has been tested on 10 CT pork carcasses and compared with manual dissection, thresholding and thresholding with bone filling techniques. We have also tested on simulated distorted images. In all the experiments our method outperforms the thresholding-based results in terms of accuracy and robustness.