Natural Image Deblurring using Recursive Deep Convolutional Neural Network (R-DbCNN) and Second-Generation Wavelets

Deconvolution in blind digital images is a common issue in image enhancement techniques, which basically was a notion of many researches. In this study, spatial varying blind deconvloution is stated and implemented. In addition, image noise removal approach which utilizes the normalized platform of second-generation wavelet transform is applied as pre-processing step. The low and high frequencies are decomposed in this step in order to be extracted. Practically, the main merit of wavelet transform is its efficiency in reduction of data redundancy in digital images. This feature helps a lot in terms of data classification where it is easy to distinguish the signal from its noisy counterpart. The second step, a recursive deep convolutional neural network (R-DbCNN) is implemented to suppress any image blur affected by second-generation wavelet transform to further remove the blur of noisy image. The experimental results depict that the suggested method outperforms recent blur removal techniques for different bluer image types in terms of image quality and time consumption.