Quality loss during storage is often associated to changes in relevant product colors and/or to the appearance of new pigments. Computer Vision System (CVS) for non-destructive quality evaluation often relies on human knowledge provided by operators to identify these relevant colors and their features. The approach described in this paper automatically identifies the most significant colors in unevenly colored products to evaluate their quality level. Its performance was compared with results obtained by exploiting human training. The new method improved quality evaluation and reduced the subjectivity and the inconsistency potentially induced by operators.

An innovative Computer Vision System (CVS) that extracts color features discriminating the quality levels occurring during fresh-cut radicchio storage in air or modified atmosphere packaging was proposed. It self-configures the parameters normally set by operators and completely automates the following steps adapting to the specific product at hand: color-chart detection, foreground extraction and color segmentation for features extraction and selection. Results proved the average value of a* 20 over the white part and the percentage of light white with respect to the whole visible surface to be the most discriminating color features to significantly separate (P <= 0.05) the three desired quality levels (high, middle and poor) occurring during fresh-cut radicchio storage 23 (whose true value was verified on the base of ammonium content and human evaluated visual quality). The proposed procedure significantly simplify the CVS design and the optimization of its performance, limiting the subjective human intervention to the minimum.