Dual Autoencoder Network with Separable Convolutional Layers for Denoising and Deblurring Images.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Image Processing

Background:

• Image noise and blur degrade visual information, necessitating effective restoration techniques.
• Traditional methods often struggle with complex noise types and parameter efficiency.
• Deep learning models, particularly autoencoders, show promise but can be computationally intensive.

Purpose of the Study:

• To develop a novel dual autoencoder architecture for simultaneous image denoising and deblurring.
• To reduce the number of trainable parameters in the neural network.
• To improve the accuracy and similarity between restored and original images.

Main Methods:

• Implementation of a dual autoencoder structure where the first autoencoder handles denoising and the second enhances the denoised image.
• Utilization of separable convolutional layers to decrease network complexity and parameter count.
• Evaluation across various noise types: Gaussian, Poisson, speckle, and random impulse noise.

Main Results:

• The proposed dual autoencoder significantly reduces trainable parameters compared to standard convolutional autoencoders.
• Achieved higher accuracy in image restoration, evidenced by decreased Mean Square Error (MSE) and increased Structural Similarity Index (SSIM).
• Demonstrated superior performance in denoising and deblurring across diverse noise conditions.

Conclusions:

• The dual autoencoder with separable convolutional layers offers an efficient and effective solution for image restoration.
• This architecture successfully balances improved image quality with reduced computational complexity.
• The findings highlight the potential of this method for practical image processing applications.