Scholarlink

<p>Multi Focus Image Fusion (MFIF) is one of the emerging areas in the field of image processing and deep learning. It means compressing images that have different focus depths into a single image which ideally preserves the properties of the source images. The fusion is performed to such a degree that the output fused image must preserve all the relevant information of the input images. The proposed method aims to enhance the fusion performance of the Laplacian pyramid fusion by making use of the Convolutional Autoencoders. The result obtained from the Laplacian fusion process is subsequently provided to the Convolutional Autoencoder (CAE). The output produced by the CAE, which is referred to as the fused result, is then utilized to compute the loss. This loss is determined by comparing the fused image with the ground truth image. The comparison aims to quantify the discrepancies between the fused image and the ground truth image, thereby allowing for an assessment of the accuracy and effectiveness of the fusion process. This two-step approach of integrating CAEs with the classical fusion technique in the proposed method provides a versatile and interpretable approach to address the challenges associated with multi-focus image fusion. Empirical assessments affirm the effectiveness of the suggested method in attaining superior fusion outcomes as evidenced by improvements in Peak Signal to Noise Ratio (PSNR), Structural Similarity Index Measure (SSIM), Entropy, Edge Intensity (EI), End Point Error (EPE) and Root Mean Square Error (RMSE).</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

PyramidCAE-Fusion: a supervised multi focus image fusion technique using laplacian pyramids and convolutional autoencoders

  • 期刊论文

  • 发表日期：

    2026年1月22日

查看全文
  • Santhi Krishna M S,
  • Harikrishnan D,
  • Shiyas C R

摘要

Multi Focus Image Fusion (MFIF) is one of the emerging areas in the field of image processing and deep learning. It means compressing images that have different focus depths into a single image which ideally preserves the properties of the source images. The fusion is performed to such a degree that the output fused image must preserve all the relevant information of the input images. The proposed method aims to enhance the fusion performance of the Laplacian pyramid fusion by making use of the Convolutional Autoencoders. The result obtained from the Laplacian fusion process is subsequently provided to the Convolutional Autoencoder (CAE). The output produced by the CAE, which is referred to as the fused result, is then utilized to compute the loss. This loss is determined by comparing the fused image with the ground truth image. The comparison aims to quantify the discrepancies between the fused image and the ground truth image, thereby allowing for an assessment of the accuracy and effectiveness of the fusion process. This two-step approach of integrating CAEs with the classical fusion technique in the proposed method provides a versatile and interpretable approach to address the challenges associated with multi-focus image fusion. Empirical assessments affirm the effectiveness of the suggested method in attaining superior fusion outcomes as evidenced by improvements in Peak Signal to Noise Ratio (PSNR), Structural Similarity Index Measure (SSIM), Entropy, Edge Intensity (EI), End Point Error (EPE) and Root Mean Square Error (RMSE).