Marbling is an important factor in evaluating pork quality and can be estimated by marbling scores based on the official marbling standards. The marbling score is normally assessed by experienced graders by comparing pork chops with the standardized chart system. In this paper, the potentials of automatic objective prediction of marbling scores were studied. The region of interest (ROI) of the marbling standards and the pork samples was automatically determined for marbling extraction. Marblings were regarded as kind of line patterns and thereby extracted by the wide line detector. Proportion of marblings (PM) was used for determinating the marbling score. The stepwise procedure was employed to select prediction models. A multiple linear regression equation was used as the initial model of the procedure and the PM of marbling standards at all three channels as potential variables. Three models were developed
by the stepwise procedure with different first entry variable of the initial model. The multiple linear
model obtained by the PM of marbling standards at all three RGB channels outperformed the two simple
linear models respectively developed at the green and blue channels. The adjusted coefficient of determination
(R2) of the multiple linear model was 0.9992 and the root mean square error of leave-one-out
cross-validation (RMSECV) was 0.0938. Forty pork loin samples were used to predict marbling scores.
The prediction results of the three models showed that the prediction ability of the simple linear model
developed at the blue channel was comparable with the multiple linear model.approach performs most robustly in comparison with the others.