Estimation and correction of motion blur in digital images

Abstract

Digital images play a very important role in developing computer-aided systems. The motion blur and blur in such types of images affect the accuracy of the system. Therefore, it is a challenging task to estimate and remove the blur in the images. In the present paper, an attempt is made to use a Convolutional Neural Network (CNN) model to estimate and remove the blur in the images. The CNN model with different functions helps to improve the accuracy of removing blur from the images. Different network functions, such as ReLU and Sigmoid, and their combinations are analyzed for the modeling of CNN. The performance of CNN is analyzed with different parameters, such as blur estimation, PSNR, RMSE, SSIM, and MSE. The performance is measured by considering different image categories, such as more blur images, less blur images, dark blur images, and biomedical images. Considering the parameters, it is observed that CNN with ReLU and Sigmoid functions is giving better performance than other network functions. It is observed that CNN models are giving successful performance to remove blur and correct the blur than any other traditional models.