Prediction of panellists' perception of bread crumb appearance using fractal and visual textural features

Ratings of visual characteristics of bread crumb images obtained by panellists were correlated with features obtained by digital fractal and texture analysis and simple thresholding. Trained panellists were asked to rate 168 bread crumb images on fineness, homogeneity and degree of orientation, using continuous line scales. The relative orientation of the main and secondary peaks of the image power spectrum was the only parameter related to the human perception of the degree of crumb orientation. Single fractal dimension terms correlated better with the panellists' perception of grain fineness and homogeneity than the single crumb features from thresholding. Second-order polynomial models were significantly better (P < 0.01) in most predictors than simple linear models. Grain fineness was better approached by the method of relative differential box-counting fractal dimension (R-2= 0.822) whereas grain homogeneity was highly related to the mass fractal dimension (R-2= 0.820). Multiple linear models to estimate grain fineness with higher predictive capacity included predictors such as fractal dimension, mean intercellular distance and void fraction (R-2> 0.860).