• Version
• Download 15
• File Size 452.34 KB
• File Count 1
• Create Date 4 June 2023
• Last Updated 4 June 2023

Image Denoising Non-Local Means Framework

Vaishnavi Ugale1, Shubham Todkar2, Siddhika Patil3 , Viraj Sure4

Department of Computer Engineering & RMD Sinhgad School of Engineering Warje,Pune-411058

---------------------------------------------------------------------***---------------------------------------------------------------------

Abstract -This Image denoising is a fundamental task in image processing, aiming to remove unwanted noise while preserving the underlying structure and details of an image. In this research paper, we propose a novel approach for image denoising by combining the Non-Local Means (NLM) framework with Convolutional Neural Networks (CNN). The NLM framework exploits the redundancy in natural images to effectively denoise by utilizing similar patches from the image. However, traditional NLM methods often suffer from high computational complexity and limited denoising performance. To address these limitations, we integrate CNN into the NLM framework to enhance its denoising capabilities.

Our proposed method first utilizes the NLM framework to estimate the similarity between patches in the noisy image. The similarity information is then used to construct a weighted average of similar patches, aiming to restore the true image structure. Next, we incorporate a CNN component to further refine the denoised image. The CNN is trained on a dataset of clean images and their corresponding noisy versions, enabling it to learn the complex patterns and structures inherent in natural images. This allows the CNN to effectively suppress the residual noise and enhance the visual quality of the denoised image.

To evaluate the performance of our approach, extensive experiments were conducted on various benchmark datasets, including both synthetic and real-world noisy images. The results demonstrate that our proposed method achieves superior denoising performance compared to traditional NLM methods and standalone CNN-based denoising approaches. The integration of NLM and CNN effectively combines the strengths of both methods, resulting in enhanced noise removal and preservation of image details.

Furthermore, our method exhibits computational efficiency, making it suitable for real-time applications. The proposed approach provides a promising solution for image denoising, offering improved visual quality and accurate restoration of the true image structure. The combination of NLM and CNN demonstrates the potential for leveraging the complementary strengths of different techniques to address the challenges in image denoising tasks.

Key Words: Non Local Mean, CNN, Mean squared error (MSE), Structural similarity index (SSIM), Gaussian Noise, Impulse Noise.