This paper considers the problem of assessing the accuracy of image registration using a multi-reference algorithm. This algorithm is based on a random selection of reference areas in the image, with the help of which geometric transformations are estimated. The similarity image transformation model is used in the work. The novelty of this article lies in the fact that, unlike other articles, methods for assessing the accuracy of image registration are proposed to improve the quality of image analysis. The correlation matrix of errors in the estimation of the parameters of geometric transformations and the correlation matrix of registration errors for the entire image have been determined. The nature of the dependence of the registration error variance on the zone of choice of reference areas is shown. Experimental studies were carried out on images obtained by semi-natural modeling based on video captured by a vehicle traffic control system. The research results can be applied when developing image processing algorithms, for example, image stabilization, registration, and moving object detection.