Detection of Distorted Meat Image for Pork Grading System

Abstract

This paper proposes a method that detects optical distorted areas (aka bubble) in pork images. By correctly identifying and discarding the images containing the unwanted bubbles, a significant improvement of pork image classification (or pork grading) in terms of accuracy has been achieved. The proposed bubble detection method relies on a particular set of image pre-processing techniques followed by morphological and region segmentation operations and is designed to attain the highest bubble detection accuracy for the detection of distorted images. Combining the proposed method with a typical pork image classification technique, the overall classification accuracy has been obtained as high as 96%.