SEMI-AUTOMATIC CRACKS CORRECTION BASED ON SEAM PROCESSING, STOCHASTIC ANALYSIS AND LEARNING PROCESS

Abstract

The restoration process of cracked images is a challenge and an important field of image and video processing. Old images that experienced bad treatment and environmental conditions, or images of old buildings and statues have the problem of cracks. This problem restricts the ability of extracting information and processing of the image. Many algorithms have been proposed to restore cracked images but most of them failed to remove these cracks efficiently based on realistic assumptions. So, we developed and implemented a new algorithm in order to repair cracked images efficiently by proposing new techniques such as seam processing, stochastic analysis and learning process. The basic motivation of this work is to design a simple algorithm of efficient computation complexity and memory usage that can be used in an interactive fashion where a simple set of parameters is used to control behavior and performance of the algorithm. The algorithm uses seam processing to discover cracks and local spatial information to compensate and handle information shortage based on statistical analysis and data generation. In general, the algorithm can be divided into three phases: cracks detection, cracks filling, and post-processing phase to enhance the quality. The results show how the algorithm deals with two cracked images of extreme deterioration. The results are based on subjective tests where 85 persons graded and evaluated the results under different conditions through three separate sessions.