Block-based noise variance estimation algorithm in blurred and noisy images with its application to motion deblurring

Abstract

The present research is organized with a focus on a block-based algorithm to estimate the image noise variance in the blurred images, in order to be applicable in the area of motion deblurring approach. The main problem in estimating the image noise variance through traditional available applications is the loss of real edge lines in the process of creating smooth areas. This complicated point is considered in the present research to be solved through double-side edge lines for the purpose of identifying the potential smooth areas. The blind algorithm presented here finds large rectangles around each end-side of edge lines. This utilizes the statistical measures to estimate the noise variance in the selected areas. It should be noted that the most important difference between the approach presented here and the available literatures, is its usage in case of blind algorithm and also the low computational complexity. It assists us to present a real-time application, in one such case. The results show that the algorithm estimates the noise variances, successfully, based on a number of standard images by developing the motion deblurring approach. Afterwards, in line with the results of the image noise variance estimation, the motion deblurring approach has been designed to retrieve the original version of the blurred and noisy image, which is under the influence of uniform linear motion with constant velocity. In order to consider the effectiveness of the proposed approach, the outcomes regarding some comparison criteria including the peak signal to noise ratio, the structural similarity, the computational complexity and finally the size of processed images have been compared with those obtained from the well-known linear motion approaches, as the potential benchmarks.