Scholarlink

<p>In this paper, we present an approach for classification of scene images based on extraction of Dense-SIFT and extended Local Directional ternary pattern (X-LDTP) feature descriptor with bag of visual words (BOVW) representation. Our approach extracts Dense-SIFT feature descriptor to capture the local context of semantic and spatial representation and eXtended-LDTP descriptor is employed to extract texture and shape information. The extracted descriptors are clustered using k-means algorithm to generate a dictionary of visual words. The visual words generated are normalized using tf-idf weighting strategy to determine the significance of visual words generated from each descriptor. Finally, an image is represented using combined tf-idf weighted histogram. SVM classifier is employed to classify images based on the similarity of feature vectors. Experiments were conducted using challenging scene image datasets such as SUN and CALTECH 256, with results being evaluated through classification accuracy. Compared with its competitors, a significant improvement of the proposed method.</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

Secne Image Classification Based on Combining Extended-Local Directional Ternary Pattern and Dense-SIFT Descriptor

  • 期刊论文

  • 发表日期：

    2025年8月7日

查看全文
  • Shrinivasa Sarathi Rangappa,
  • Prabhakar Chikkanavangala Jambanna

摘要

In this paper, we present an approach for classification of scene images based on extraction of Dense-SIFT and extended Local Directional ternary pattern (X-LDTP) feature descriptor with bag of visual words (BOVW) representation. Our approach extracts Dense-SIFT feature descriptor to capture the local context of semantic and spatial representation and eXtended-LDTP descriptor is employed to extract texture and shape information. The extracted descriptors are clustered using k-means algorithm to generate a dictionary of visual words. The visual words generated are normalized using tf-idf weighting strategy to determine the significance of visual words generated from each descriptor. Finally, an image is represented using combined tf-idf weighted histogram. SVM classifier is employed to classify images based on the similarity of feature vectors. Experiments were conducted using challenging scene image datasets such as SUN and CALTECH 256, with results being evaluated through classification accuracy. Compared with its competitors, a significant improvement of the proposed method.