USING OF MULTI-DIMENSIONAL CLASSIFICATION METHODS FOR DETERMINING FRAGMENTALLY DISTORTED IMAGES

Abstract

The paper proposes a method for determining distorted images based onthe use of classification methods for data analysis. A series of digital images modified byinserts with pixel mixing were prepared for research. It is proposed to use local histogramsas informative parameters characterizing the distortion of brightness in the image area. Toestablish the differences in the informative parameters of unmodified and distorted images,the classification methods PLS-DA, SVM, the method of k-nearest neighbors were used.Comparison of the classification results made it possible to establish that the use of theSVM classifier makes it possible to more accurately determine the classes of unmodifiedand distorted images.