Multi-View Image Denoising Using Convolutional Neural Network.

Area of Science:

• Computer Vision
• Image Processing
• Machine Learning

Background:

• Multi-view images present unique denoising challenges due to varying disparities.
• Conventional denoising methods often involve computationally intensive patch searching.

Purpose of the Study:

• To propose a novel and efficient multi-view image denoising algorithm.
• To leverage convolutional neural networks (CNNs) for improved denoising performance.

Main Methods:

• Arranging multi-view images into 3D focus image stacks (3DFIS) based on disparities.
• Employing a multi-view convolutional neural network (MVCNN) trained on 3DFIS.
• Utilizing residual learning and batch normalization within the MVCNN architecture.
• Fusing denoised image stacks using estimated disparity maps to generate the final denoised image.

Main Results:

• The MVCNN algorithm effectively denoises multi-view images with Gaussian noise.
• The proposed method significantly reduces computational time compared to traditional patch-based approaches.
• Experimental results demonstrate high effectiveness and efficiency against state-of-the-art methods.

Conclusions:

• The MVCNN offers a highly effective and efficient solution for multi-view image denoising.
• The CNN-based approach eliminates the need for exhaustive patch searching, streamlining the denoising process.