An Effective Image Feature Selection and Mining Algorithm

Abstract

Feature selection is a key problem in image recognition, the number of original features is very large. The image characteristics are of small sample, high dimension, great noisy, high redundancy and nonlinear, the linear correlation analysis can partly show the image feature data rules and the simple linear structure of data, it can’t show the nonlinear nature and complexity. The paper proposes a feature selection and mining algorithm through high step association analysis, mines the various features of object recognition using property related analysis, the distinguishing bridge features cover energy, homogeneity, contrast and the regional mean ratio characteristics, it makes up the defects that linear correlation can not accurately determine the nonlinear structure. The sensitive recognition features are chosen to improve the recognition efficiency. The experiment results show the algorithm has effectively improved the accuracy of bridge recognition.