Identification of Bruised Apples Using a 3-D Multi-Order Local Binary Patterns Based Feature Extraction Algorithm

In this paper, we propose an algorithm for identifying bruised apples based on 3-D shape information obtained by a 3-D infrared imaging system. The algorithm aims at classifying the harvested apples into two classes: bruised apples and un-bruised apples. The proposed algorithm is composed of two steps: feature extraction and classification. For feature extraction, we introduce a vertex-based mesh local binary pattern (vmLBP) operator to extract binary patterns from 3-D meshes. Specifically, we design a framework to construct ordered vertex rings (OVRs) for the computation of vmLBP codes. The constructed OVRs are designed to have the same ordering fashion as each other. For classification, we apply a support vector machine classifier to train the feature vectors generated from the histograms of vmLBP codes. Through experiments, we investigated and optimized the parameters of the proposed algorithm to achieve the highest identification accuracy. We compared the proposed algorithm with other algorithms. Experimental results show that the proposed algorithm has achieved a better performance for bruised apple identification than traditional algorithms, which indicates that the proposed vmLBP operator has good potential in a 3-D shape analysis due to its high discriminative power.