Multidirectional Gradient Neighbourhood-Weighted Image Sharpness Evaluation Algorithm

Abstract

Aiming at the problem that the image sharpness evaluation algorithm in the photoelectric system has a slow speed in actual processing and is severely disturbed by noise, an improved image sharpness evaluation algorithm is proposed by combining multiscale decomposition tools and multidirectional gradient neighbourhood weighting. This paper applies non-subsampled shearlet transform (NSST) to perform multiscale transformation of the input images, obtaining high-frequency sub-band images and low-frequency sub-band images. In order to enhance the detection of the edge orientation of images, multidirectional gradient processing of the image matrix is added to each sub-band image. In addition, the weight corresponding to the current pixel is obtained by calculating the inverse ratio of the gradient of each direction and the distance of the center pixel. Through calculating the ratio of the gradient neighbourhood weighting operators of high-frequency sub-band images and low-frequency sub-band images, the image sharpness evaluation value can be acquired further. Moreover, the image sequence collected by a certain type of photoelectric system is selected as the image sequence of the noisy real environment for simulation experiments and compared with the current mainstream algorithms. Finally, the experimental draws a conclusion that compared with the mainstream evaluation algorithms, the evaluation results of the proposed method perform better in terms of steepness, sensitivity, and flat area fluctuation, it can better suppress noise and improve accuracy, and its running speed meets the basic requirements of the image sharpness evaluation algorithm.

Improved SMD Image Evaluation Function Based on Pixel Difference

“…Sometimes the calculated result has a large error, causing mis-adjustment. The subjective evaluation habit of the human eye is to compare the center pixel with the adjacent four or eight pixels [9].…”