QUANTITATIVE EVALUATION OF WIENER’S IMAGE FILTERING

Abstract

When processing and transmitting information on the image, the problem of reducing their distortion due to various noises is relevant. Noise reduces the quality of the image and, accordingly, the perception of the information contained in it. This, in particular, reduces the ability to evaluate the information that can be obtained as a result of analysis using both visual and computer methods. The procedure of reducing noise in images is handled by an area of image processing called restoration. Despite the intersection of this area with image enhancement, it should be noted that the latter is more of a subjective procedure, while the restoration process is objective. During restoration, an attempt is made to reconstruct or reproduce the distorted image, using a prior information about the occurrence that caused its deterioration. Restoration methods are based on the modeling of distortion processes and the use of reverse procedures to restore the original image. In this work, a method of quantitative assessment of restoration of distorted images using Wiener filters is advanced. It consists in obtaining histograms of distorted images and their extrapolation with a Gaussian curve, followed by determination of the value of the mean-square deviation. A similar procedure is carried out for the restored image. It is proposed to carry out a quantitative assessment of the degree of restoration using the parameter R, which determines the relative narrowing of the histogram of the restored image in relation to the distorted one. Such estimates were made for the Wiener filter. Also, for comparison, the median filter and its combined effect with the Wiener filter were used. The best result was obtained with the sequential action of the Wiener and median filters, which was confirmed both by the maximum value of the R parameter and by visual observations. The analysis was performed in the Python programming language using the Pillow and OpenCV image libraries.

Keywords: image restoration, Gaussian noise, Wiener filter, median filter, image restoration parameter, Python.