Abstract <p>The article addresses the problem of assessing the boundaries of segmentation quality indicators: variation of information and normalized variation of information between the original image and its segmentation maps, as well as the segmentation error probability‒mutual information function. We propose an information segmentation model, in which image transformations are considered as information channels. We estimate the lower and upper bounds of the quality indicators, which are calculated as functions of the segmentation error probability for the specified information characteristics of the input image and its ground truth partition. Examples of constructing boundaries for test images are given. The constructed boundaries make it possible to estimate the range of possible values of segmentation quality measures for a given original image and its reference partition without performing segmentation. These bounds can be used to estimate the quality of existing ground truth partitions and select the most appropriate one to the problem being solved. The proposed bounds can be applied both to select a segmentation algorithm and to adjust the parameters of the selected algorithm.</p>