Denoising infrared maritime imagery using tailored dictionaries via modified K-SVD algorithm

Area of Science:

• Image Processing
• Computer Vision
• Signal Processing

Background:

• Tailored overcomplete dictionaries offer superior image modeling compared to standard basis functions.
• Existing dictionary learning algorithms like K-SVD are effective but can be computationally intensive.

Purpose of the Study:

• To introduce a modified K-SVD algorithm for enhanced dictionary learning.
• To improve convergence speed while retaining the benefits of the original K-SVD approach.
• To apply the learned dictionary for denoising infrared maritime imagery.

Main Methods:

• Development of a modified K-SVD dictionary learning algorithm.
• Denoising of infrared maritime imagery using the learned dictionary.
• Comparative analysis against the original K-SVD, fixed overcomplete dictionaries, and wavelet denoising.

Main Results:

• The modified K-SVD algorithm demonstrates improved convergence.
• Overcomplete representations are shown to be superior for image denoising.
• The tailored approach achieves similar Peak Signal-to-Noise Ratios (PSNR) to traditional K-SVD.
• The proposed method achieves these results at approximately half the computational cost.

Conclusions:

• The modified K-SVD algorithm provides an efficient and effective method for learning overcomplete dictionaries.
• This tailored approach enhances the denoising of infrared maritime imagery.
• The study highlights the advantages of customized dictionary learning for specialized image processing tasks.