Nonlinear filtering in digtal image processing

Abstract

In this thesis an application named ImageLab have been developed. Image-Lab is an image filtering application focusing on the use of nonlinear filteringprocesses. During development of the software it became clear that ImageLabis not a full image processing suite, such as Photoshop , but rather a complementto existing software solutions. This is true for now, but further developmentof the software may lead it to become an alternative full featured image processingsolution.Theory behind the use of ImageLab s image processing techniques was introducedin section 2, the same techniques was also introduced as pseudocode.Also the different filters and their functionality were explained. In section4.2 the same pseudo-code were turned into usable C#-code. Following thisrecipe it should be a straightforward process to duplicate the work done in thisthesis, both for novel applications or extending the current application further.In section 5 it was shown how the different filters performed with respect toan objective metric. While not being able to generally conclude a given filtersproficiency at removing noise, we can conclude that for a given set of initial conditions (input image, the type/strength of noise and parameters) some filters didperform better than others. With 2-sided variable height impulsive noise andthe Lena reference image a median-filter and a Distance Weighted Median-filterperformed the best (according to the objective Mean Square Error). All the filtersdid improve the objective metric, the worst (Mean 11-by-11) was able toremove more than 40%4 of the noise. The best filter (Median 1-by-1) was ableto remove more than 90%5 of the noise.