Color of Salmon Fillets By Computer Vision and Sensory Panel

A computer vision method was developed and used to assign color score in salmon fillet according to SalmonFan (TM) card. The methodology was based on the transformation of RGB to L*a*b* color space. In the algorithm, RGB values assigned directly to each pixel by the camera in the salmon fillet image, were transformed to L*a*b* values, and then matched with other L*a*b* values that represent a SalmonFan score (between 20 and 34). Colors were measured by a computer vision system (CVS) and a sensorial panel (eight panelists) under the same illumination conditions in ten independent sets of experiments. Errors from transformation of RGB to L*a*b* values by the CVS were 2.7%, 1%, and 1.7%, respectively, with a general error range of 1.83%. The coefficient of correlation between the SalmonFan score assigned by computer vision and the sensory panel was 0.95. Statistical analysis using t test was performed and showed that there were no differences in the measurements of the SalmonFan score between both methods (t (c) = 1.65 a parts per thousand currency sign t = 1.96 at alpha = 0.05%). The methodology presented in this paper is very versatile and can potentially be used by computer-based vision systems in order to qualify salmon fillets based on color according to the SalmonFan card.