Deep convolutional neural network based image de-noising with wavelet transform

• Main Authors: Dixit, Madhuvan, Pawar, Mahesh
• Format: Conference Proceeding
• Language: English
• Subjects: Artificial neural networks; Image degradation; Image filters; Image quality; Image restoration; Noise measurement; Noise reduction; Qualitative analysis; Random noise; Signal to noise ratio; Smoothness; Structural members; Visual aspects; Wavelet transforms
• ISSN: 0094-243X; 1551-7616
• DOI: 10.1063/5.0234317

One important element that degrades the quality and look of a picture when introduced is noise. Therefore, it has to be removed without affecting the image’s structural elements or textural information in order to increase its quality. Sounds come in many different ways that might spoil pictures. The application determines which de-noising method is used. Research has shown that Deep CNN and Feature Transform based image denoising performs better than traditional vision-based methods. Nevertheless, there are also a few limitations that lead to additional distortions or are only partially successful in recovering visual characteristics like textured areas. The quality of the picture falls under qualitative analysis, taking into account the edge factor, texture, smoothness, uniform and non-uniform regions, and object structure. Quantitative research compares the results of a DCNN-FT based method with those of a traditional or standard denoising method using the metrics of Structure Similarity Index Measurement (SSIM) and Peak Signal to Noise Ratio (PSNR). This may be a prerequisite for learning the denoising procedure. It has been verified that compared to previous single-input networks, our suggested multi-input network can produce superior denoising performance. The research and testing findings demonstrate that, in comparison to other conventional/standard image filtering methods, the DCNN-FT model is capable of effectively removing a large amount of Gaussian noise and restoring the picture features and data.