Training image-correlation systems by optimizing their attribute representations

Аверкин А.Н. Обучение систем сопоставления изображений путем оптимизации их признаковых представлений // Оптический журнал. 2010. Т. 77. № 11. С. 69–79.

Averkin A.N. Training image-correlation systems by optimizing their attribute representations [in Russian] // Opticheskii Zhurnal. 2010. V. 77. № 11. P. 69–79.

A. N. Averkin, "Training image-correlation systems by optimizing their attribute representations," Journal of Optical Technology. 77(11), 712-720 (2010). https://doi.org/10.1364/JOT.77.000712

A new method has been developed for representing the key points of images by vectors in attribute space. The method is based on an algorithm for the adaptive construction of new attributes on the basis of descriptions obtained ahead of time by using the “speeded-up robust features” method for a series of images recorded indoors from different aspects. Experimental studies of the model thus developed on various series of images shows that this method can be used in practice to solve the image-correlation problem, provided that a preliminary training stage of the given model is carried out.