Scholarlink

<p>This study proposes a similarity index (SI) to evaluate the overlap between two probability density functions (PDFs), which has proven effective for assessing distributional divergence compared to existing metrics, particularly in multivariate datasets. The specific for SI is presented, the properties are investigated, and its relationship with other common measures is also established. The SI is then applied to improve image classification algorithm when the pixel features of each image are extracted and constructed into a representative PDF. Based on the SI, the proposed algorithm improves the method of finding prior probabilities and the quasi-Bayesian classification principle. An image is assigned to a group if it has the highest prior probability and SI to that group. By incorporating these improvements, we obtain an effective classification algorithm for images. The proposed algorithm is detailed the implementation steps, considered the convergence, and performed by the established Matlab procedure. With the considered 6 real datasets, the proposed algorithm has provided stable, high accuracy, outstanding, and competitive classification results with many statistical, machine learning, and deep learning classification methods.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Developing an image classification algorithm based on the novel measure combined with quasi-Bayesian method for representing probability density function

  • 期刊论文

  • 发表日期：

    2025年9月23日

查看全文
  • Tai Vo-Van

摘要

This study proposes a similarity index (SI) to evaluate the overlap between two probability density functions (PDFs), which has proven effective for assessing distributional divergence compared to existing metrics, particularly in multivariate datasets. The specific for SI is presented, the properties are investigated, and its relationship with other common measures is also established. The SI is then applied to improve image classification algorithm when the pixel features of each image are extracted and constructed into a representative PDF. Based on the SI, the proposed algorithm improves the method of finding prior probabilities and the quasi-Bayesian classification principle. An image is assigned to a group if it has the highest prior probability and SI to that group. By incorporating these improvements, we obtain an effective classification algorithm for images. The proposed algorithm is detailed the implementation steps, considered the convergence, and performed by the established Matlab procedure. With the considered 6 real datasets, the proposed algorithm has provided stable, high accuracy, outstanding, and competitive classification results with many statistical, machine learning, and deep learning classification methods.