Sparse and Redundant Image Representations Using Adaptive Dictionaries in Digital Image Denoising

Digital image denoising is a widely know problem in image processing. In this work, we focus on removing additive white Gaussian noise from a given image. We use a denoising method that was extended from the Sparseland model. This method builds sparse image patch representations using redundant dictionaries and has been shown to denoise images fairly well. We look to improve this method by adapting the dictionaries to more accurately represent specific image features. The image features were chosen to be the details, textures, and smooth regions of the image. Two different dictionaries were tested, the Discrete Cosine Transform and a learned dictionary based off of the noisy image patches. The dictionary's patch size was also varied to find the optimal patch size for denoising each image feature. Numerical and visual comparisons show the promise of this method improvement.

Name thesis:degree_name

MS

Level thesis:degree_level

Immediate Access

Discipline thesis:degree_discipline

Computational Mathematics

Year dc:date.available

2009

Author dc:creator

• Sano, Teresa

Contributors dc:contributor

• Stacey Levine
• Carl Toews
• Mark Mazur

• image processing
• denoising
• sparsity
• geometric features

Language dc:language

English

Repository record dc:identifier

https://dsc.duq.edu/etd/1145

OAI identifier oai:identifier

oai:dsc.duq.edu:etd-2161