An image processing method for in-line nectarine variety verification based on the comparison of skin feature histogram vectors

• The proposal of an image processing method for in-line nectarine variety verification is proposed.
• The use of feature vectors from the histogram raw values of different color planes are studied.
• The definition of a variety classification threshold that determines the similarity between nectarines.
• The conclusion that a feature histogram vector created of the red and gray color layers enables nectarine variety validation.
• The conclusion that the proposed system has better classification performances than a human operator.

This paper presents an image processing method for in-line automatic and individual nectarine variety verification in a fruit-packing line based on the use of feature histogram vectors obtained by concatenating the histograms computed from different color layers of a circular central area of the skin of the nectarines processed. The verification procedure requires the definition of a small dataset with the feature histogram vectors corresponding to some reference nectarines (manually selected) whose skin clearly identifies the variety being processed. The in-line variety verification of each nectarine processed is then done by computing and comparing its current feature histogram vector with the reference dataset. This paper compares experimentally different alternatives for computing the feature histogram vectors and two methods for feature comparison and variety verification. The experimental validation consists of the automatic in-line processing of nectarine samples from different mixed varieties. The results show an 86% success rate in the case of an expert human operator and 100% when using feature histogram vectors computed in the Rg   (red and gray) or YR‾ (luminance and normalized red) intensity color layers and when using correlation to compare the feature vectors.