Active contour-based segmentation of coffee cherries images

A coffee-fruit (cherry) image segmentation system that uses a fuzzy active contour approach is presented. Images processed contain only fruits at any ripeness stage, which are detected processing the clusters given by the fuzzy c-means algorithm, applied to the RGB color space. The segmentation is performed using a fuzzy active contour procedure, composed by an iterative process which starts from the convex hull of a seed region and compute, at each iteration i, the coffee affinity of the i^th contour. The affinity measurement uses shape descriptors. The spectral information is used to find the seed region and to compute the initial energy of its convex hull (seed contour). A stop criterion defines when the iterative process is finished and, later, the best contour which defines the coffee fruit is found. An 70% of correct detection is reached, and this performance is compared with other systems that use a region growing approach.