Segmentation of AFM images based on wave regions growing of local maximums with pixels selection in the decrease order of values

Abstract

An algorithm for segmentation of images of atomic force microscopy is developed by using wave-growing regions around local maxima as a result of adjoining neighboring pixels to them, selected in descending order of values. The essence of the algorithm is to use the brightness threshold, gradually changing from maximum to minimum, to select growth points or to join existing areas. The features of the developed segmentation algorithm are iteratively expandable boundaries, the choice of initial growth points and points attached to regions with a focus on threshold values with a gradual decrease from maximum to minimum. These features made it possible to eliminate the segmentation errors of the AFM images, characteristic of the algorithms of the marker watershed, the cultivation of areas and the watershed using the classical Vincent–Sollie algorithm, which are commonly used. The developed algorithm was compared with standard algorithms such as: classic watershed algorithm, marker watershed, growing areas. The comparison was carried out on test and original AFM images. The algorithms were implemented in Matlab and C ++. A set of binary masks was used to quantify segmentation errors. The results showed that the developed algorithm provides the selection of the boundaries of the regions without errors and a higher segmentation rate in comparison with the algorithms of growing the regions and the Vincent–Sollie watershed. The result can be used to process AFM images of the surfaces of inorganic materials in the submicro and nano range.