Scholarlink

This work provides a thorough examination of developments in medical image analysis, with an emphasis on image preprocessing methods to improve raw images in preparation for further tasks including object identification, extraction of features, and classification The study delves into image augmentation, employing techniques like zoom, shear, rotation, and flips, including horizontal and vertical flips, and introduces grid distortion for dataset augmentation. Beyond preprocessing, it explores image transformation operations, encompassing resizing, pixel value normalization, and histogram equalization, coupled with geometric transformations to enhance dataset robustness. The investigation delves into image compression techniques, specifically analyzing the Haar wavelet transform and discrete cosine transform (DCT) for both lossless and lossy compression. Subsequently, this study explores segmentation models (U-Net, DeepLabV3+, and Iterative U-Net) and classification models (Random Forest, EfficientNet B0, and an Ensemble model). The methodology outlines the research approach, including dataset acquisition, image preprocessing, segmentation, and classification, with a detailed performance analysis. The results highlight the superior accuracy of the Iterative U-Net and Ensemble model, reaching 92.27% on DRIVE and EyePACS dataset. This study concludes by emphasizing the crucial role of medical image segmentation, suggesting future avenues for exploration, and serving as a foundation for continuous innovation in medical image analysis and diagnosis.

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

Enhancing Segmentation Accuracy of Diabetic Retinal Images Using Transform Techniques and deepLabV3+ with Ensemble Classification

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • S. Mohana Saranya,
  • T. Pranesh,
  • R. Rajasurya,
  • A. H. Sabareesh,
  • Dinesh Komarasamy,
  • S. Mohanapriya

摘要

This work provides a thorough examination of developments in medical image analysis, with an emphasis on image preprocessing methods to improve raw images in preparation for further tasks including object identification, extraction of features, and classification The study delves into image augmentation, employing techniques like zoom, shear, rotation, and flips, including horizontal and vertical flips, and introduces grid distortion for dataset augmentation. Beyond preprocessing, it explores image transformation operations, encompassing resizing, pixel value normalization, and histogram equalization, coupled with geometric transformations to enhance dataset robustness. The investigation delves into image compression techniques, specifically analyzing the Haar wavelet transform and discrete cosine transform (DCT) for both lossless and lossy compression. Subsequently, this study explores segmentation models (U-Net, DeepLabV3+, and Iterative U-Net) and classification models (Random Forest, EfficientNet B0, and an Ensemble model). The methodology outlines the research approach, including dataset acquisition, image preprocessing, segmentation, and classification, with a detailed performance analysis. The results highlight the superior accuracy of the Iterative U-Net and Ensemble model, reaching 92.27% on DRIVE and EyePACS dataset. This study concludes by emphasizing the crucial role of medical image segmentation, suggesting future avenues for exploration, and serving as a foundation for continuous innovation in medical image analysis and diagnosis.