Binarization of "background–object" by using pretreatment of an image and fuzzy $k$-means algorithm

The paper considers the stage of binarization of color $2\mathrm{D}$-image into the connected areas with the pixels of an object and background while identifying the state of the objects of natural origin and their mass quantity with detailed evaluation by means of the systems of computer vision, with high variability inside the classes and their proximity. Binarization algorithm is suggested on the basis of $k$-means clusterization with Euclid’s metrics, which helps eliminate a number of difficulties while choosing the algorithms for software actualization in the computer vision systems and satisfies the demands for velocity and accuracy. The algorithm combines pretreatment of an image by using additional "fuzzy" cluster in the $k$-means algorithm for the control of quality of separation of an object from background. The pretreatment under certain conditions excludes the lightness $L$ of color scale CIELab (CIE 1976 Lab) from $k$-means.