Аннотация:We propose a method of color adaptation for images of polished sections of geological specimens intended to be used in mineral segmentation with a pretrained neural network model. The algorithm approximates color and brightness characteristics of the images by the corresponding values on photographs of polished sections previously used in neural-model training. The key idea of the method is choosing a transition matrix between image color spaces. The color correction algorithm is very fast and trainable on small data sample. The experimental results show that color adaptation of images by the proposed method essentially improves mineral segmentation quality relative to the basic model without color adaptation. The method can be applied to automate geological specimen analysis.