ROBUST LEAN TISSUE SEGMENTATION FOR BEEF QUALITY GRADING

Abstract

To automatically evaluate beef quality, a system must be developed to properly segment lean tissue in a sectional image of the 13th beef rib. To this end, a mobile color computer vision system and its corresponding image processing algorithms were developed for on-site application. The algorithms implement Renyi entropy and a texture index to provide adaptive thresholding. Automatic smoothing and modification followed boundary extraction, and binary morphological approaches were also taken. When 54 images of beef cut samples were assessed without manual intervention, the proposed algorithms exhibited an average boundary extraction error of 3% and an average pixel distance error of 1.8 pixels relative to assessments made by a human expert. The computing time for these sample images was of approximately 5.5 s. In addition, a user-friendly interactive man-machine interface was also developed to allow for a human expert to modin) the extracted boundary.