Detection of Fruits in Natural Background (Part 2)

Mathematical Shape Description and Fruits Searching

Abstract

Using the technique of color-ratios (R/G, R/B and G/B) thresholding and Maximum and Minimum filters to obtain the boundaries of papaya fruit (Carica papaya L.) in natural background image was proposed in Part 1. However, the result did not give a complete closed-curve boundary, because of many natural environment factors, such as overlapping and noise. To detect the size and location of the fruits with incomplete boundary image, we first determine the representative shape of the fruit and then deform artificially the representative shape in many possible spatial frequency domains. Then, the multiple shapes that match the deformation at each set of searching points are clustered by evaluating the distance between the centroids. The size and location of the fruit can be evaluated by average shapes in the same group.