Visual dictionary construction method based on Dempster-Shafer (D-S) evidence theory

Abstract

The invention discloses a visual dictionary construction method based on a Dempster-Shafer (D-S) evidence theory. The visual dictionary construction method comprises S1, abstracting scale invariant feature transform (SIFT) features of all training images, utilizing K mean to achieve primary visual dictionary classification to obtain K subclasses; S2, setting a classification threshold t and an entropy threshold s; S3, judging whether a current classification number is smaller than t, if the classification number is smaller than t, step S4 is actuated, otherwise, stepS7 is actuated; S4, calculating information entropy of each subclass according to existing classification; S5, selecting a subclass with the largest information entropy to serve as the class to be decomposed, judging whether the information entropy of the subclass is larger than s, if the information entropy of the subclass is larger than s, actuating step S6, otherwise, actuating step S7; S6, further classifying a subclass h with the largest information entropy by utilizing the D-S evidence theory; S7, calculating a cluster center of a newly formed subclass, and forming a visual dictionary. The visual dictionary construction method based on the D-S evidence theory can construct a more effective visual dictionary and improve classification accuracy of images.

Description

technical field [0001] The invention relates to the technical fields of image feature extraction, visual clustering and image classification, in particular to a method for constructing a visual dictionary based on D-S evidence theory. Background technique [0002] Image classification is a key research issue in the field of computer vision. It can distinguish different types of objects and has a wide range of applications in satellite remote sensing, aerospace, biomedicine, etc. In recent years, the method of using the bag-of-words model to represent images and realize image object classification has received greater attention in image classification because it realizes an image representation model based on multiple image features. In the bag of words model, it is necessary to first construct a visual dictionary of image classes. The quality of visual dictionary construction directly affects the effect of image classification. The current traditional method of constructin...

AI Technical Summary

Problems solved by technology

This method only considers the visual similarity of features in the process of visual clustering, which leads to ignoring the influence of different features on the construction of visual dictionary.

Images