Prediction of sensory properties of low-fat yoghurt and cream cheese from surface images

Abstract

The sensory properties of 25 plain yoghurts and 18 low-fat cream cheeses were investigated by descriptive analysis. In parallel with sensory analysis digital images of sample surfaces were captured and the relationship between image properties and sensory properties were investigated. Global image features of the yoghurt and cream cheese surfaces were extracted using the Angle Measure Technique (AMT). Multivariate data analyses (Partial Least Squares Regression) were applied for investigation of the relation between digital image global features and sensory properties. For both product categories all sensory properties could be predicted with Root Mean Square Error of Cross Validation (RMSECV) for the yoghurts in the range [1.00; 1.97] and for the cream cheeses [0.29; 2.80]. For yoghurts, the largest RMSECV is for the prediction of Creaminess, indicating that it is not well predicted from structure alone. Due to covariation with appearance and structure, sensory properties relating to different modalities (taste and flavour) could also be predicted.