Image Restoration by Graduated Non-convex Local Adaptive Priors: An Energy Minimization Approach
| Autor: | H N Latha, Rajiv R. Sahay |
|---|---|
| Rok vydání: | 2021 |
| Předmět: |
Point spread function
Blind deconvolution Markov random field Pixel Computer science ComputingMethodologies_IMAGEPROCESSINGANDCOMPUTERVISION Regularization (mathematics) symbols.namesake Additive white Gaussian noise Computer Science::Computer Vision and Pattern Recognition symbols Gradient descent Algorithm Image restoration |
| Zdroj: | Inventive Systems and Control ISBN: 9789811613944 |
| DOI: | 10.1007/978-981-16-1395-1_26 |
| Popis: | This work proposes a novel algorithm to estimate latent focused image of a degraded picture by a sparsity-based optimization framework using graduated non-convex discontinuity adaptive Markov random field (GNC-DAMRF) regularization prior by minimizing the energy term. Initially, an adaptive non-local pixels-based filtering technique is exploited to denoise the 2D picture affected by additive white Gaussian noise (AWGN). The space non-uniform blind deconvolution is developed to deblur the picture by sparsity-based technique followed by adaptive maximum-a-posterior (MAP) GNC-DAMRF regularization framework. The non-uniform space-variant point spread function (PSF) of every pixel is estimated by a method proposed by just noticeable blur (JNB) [1]. The estimation of the image and the sigma blur-map of point spread function (PSF) were focused. The proposed alternating energy minimization objective function based on the sparsity approach is with-success optimized with the generic gradient descent technique. The performance of the adaptive MAP-GNC-DAMRF work is enhanced, which is superior to Gaussian Markov random field (GMRF) regularizer and few state-of-the-art works. The reconstruction results are promising and better image qualitative assessment is obtained by PSNR and SSIM. This proposed algorithm is also effectively utilized for image restoration applications such as image in-painting, depth estimation, image denoising, super-resolution, and haze removal. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|