An adaptive threshold level for segmentation of grayscale images using image processing to sort mangoes by weight estimation.
Mata M. E. R. M. C., Duarte M. E. M., Borges F. F., Barbacena I. L.
Author Affiliation: Agricultural Engineering Department, Federal University of Campina Grande, Campina Grande, PB, Brazil.
: 943-948
Abstract : Inside a packing house, tropical fruits need to be sorted mainly according to their weight, normally using complex and expensive dynamic weighing systems. The development of a low cost system to be used by small producers, from developing countries, is still a challenge, since the sorting involves weighing while the fruit is moving. This work proposes a new threshold for segmentation of white background grayscale images of mangoes (Mangifera indica L., 'Tommy Atkins'), improving the Otsu's Method used in MATLABTM. 96 mangoes were acquired at the local market in Campina Grande (PB), Brazil. A commercial digital color camera (SonyTM P7) was fixed at the center top of a closed polystyrene box (50×50×50 cm), all internal walls covered with white paper. Two PL (20 W) lamps, placed beside the camera and pointed down, illuminated the mango, manually placed at the center bottom of the box. Top view pictures of the fruit, at rest, were taken. Offline Matlab-based software calculated the top area of the fruit using a new threshold formulae: y=x+x*(m/k), where y: new grayscale threshold value; x: Otsu's threshold; m: mean intensity of the image previously segmented using the Otsu's traditional algorithm; and k: experimental constant, light intensity and setup dependent. Results showed low correlation (segmented area × actual mass) when using the default algorithm (R2=0.68, SD=48.5 g). The alternative proposed algorithm greatly improved that correlation to R2=0.94, SD=20.5 g. In this controlled setup, better segmentation of the images with white background and better mass estimation of mangoes, individually and at rest, are possible when using a threshold based on an adaptive increased percentage of the automatic threshold used in Matlab. The results also indicate the possibility of using image processing for low cost sorting of mangoes and other fruits dynamically, on a small conveyor belt.