Adaptive Relief Feature Evaluation and Selection based on Grey Level Co-occurrence Matrix

摘要

In image recognition, how to select informative features from the feature space is a very significant task. Relief algorithm is considered as one of the most successful methods for evaluating the quality of features. In this paper, it firstly provides a valid proof which demonstrates a blind selection problem in the previous Relief algorithm. And then this paper proposes an adaptive Relief (A-Relief) algorithm to alleviate the deficiencies of Relief by dividing the instance set adaptively. Lastly, it uses grey level co-occurrence matrix (GLCM) to extract text features and applies A-Relief algorithm to classify these features. The experimental results illustrate A-Relief algorithm proposed in this paper can improve the accuracy of the classification effectively and solve the blind selection problem.