Lamb carcass classification system based on computer vision. Part 1, Texture features and discriminant analysis

Abstract

This paper presents a lamb carcass classification system based on image and texture analyses together with multivariate statistical techniques (principal component analysis, cluster analysis and discriminant function analysis). Texture analysis is based on grey level co-occurrence matrix. A set of ninety texture features has been used to extract the texture information from the acquired images. In addition, twelve image area and thickness (geometric) variables have also been calculated. Principal component analysis was used to reduce the dimensionality of the original data set. Two feature sets were generated based on the results. These feature sets comprised of principal component (PC) scores calculated from the original variables and 14 (6 geometric and 8 texture) variables selected from the original set of variables. Both feature spaces were used for discriminant analysis. From the experimental results, it was established that the system enabled 66.3% and 76.9% overall classification based on 6 geometric PC scores and 14 (geometric and texture) PC scores, respectively. The system also enabled 64% and 79 % overall classification of lamb carcasses based on 6 geometric and 14 (geometric and texture) variables, respectively. This study shows the predictive potential of combining image analysis with texture analysis for lamb grading. The addition of carcass weight improved the overall classification accuracy, of both feature sets, to 85%.