The Analytic Hierarchy Method-Based Algorithm for Restoring Broken Pixels on the Noisy Images

• Published in: Nauka i obrazovanie (Moskva) Vol. 14; no. 11; pp. 521 - 534
• Main Authors: Belim, Sergey, Seliverstov, Sergey
• Format: Journal Article
• Language: English; Russian
• Published: MGTU im. N.È. Baumana 03.11.2014
• Subjects: data recovery; impulse noise; noisy images
• ISSN: 1994-0408; 1994-0408
• DOI: 10.7463/1114.0742145

This article presents an algorithm for restoring broken pixels, which can appear in graphic files with statistical gaps. The suggested algorithm is based on the method of hierarchical analysis of the decision support theory. The choice of the broken pixel color depends on the nearest neighbors and their next neighbors. Three parameters inherent in each nearest neighbor are analysed. Firstly, it is the number of neighbors, which have the same color as the given nearest neighbor. Secondly, it is a deviation of the given pixel from the average value of its neighbors. Thirdly, it is a difference between the pixels being on the opposite sides of the broken pixel. Based on these three criteria, for each nearest neighbor of the broken pixel, a weight coefficient is defined. A hierarchical two-level tree for making decision is constructed. As a color of the broken pixel, its neighbor color with the maximum weight is chosen.A computer experiment to determine the effectiveness of the proposed method is conducted. The effectiveness of the proposed method was determined by comparing the similarity degree of the broken and restored images to the source image. To compare images Minkowski metric was used. To conduct experiments photographic and artificial images were used. The paper investigates a dependence of the proposed algorithm efficiency on the broken image amount. It was found out that the proposed algorithm has the advantage over the known algorithms for restoring broken pixels near the sharp edges. An image restored by our method has more sharply defined edges as compared to what the smoothing filters provide. The proposed method can be iteratively applied. As the experiments have shown, the first five iterations provide image enhancement.The proposed method together with the algorithms for detecting the broken pixels can be used to design filters of noisy images. The method efficiency enhancement can be achieved in case of taking into account the large number of neighbors in making decision. This method can be also developed using the increased number of criteria when making decisions.