Scale space model of image degradation

Mini Project Report, Mphill CS

Mphill. Computer Science 2014-2015 INT Joseph Suresh Paul

Degradation reduces structural information of the image and results in poor visual quality. Degradation is generally model using linear position invariant filter. Blur produced by degradation is estimated using sharpening filters to enhance the fine details. Effectiveness of the filtering technique depends on the extent and accuracy of degradation. However, in practical scenario, the extent of degradation is unknown. Estimation of the extent of degradation, therefore, would be useful to understand and validate different image sharpening methods. In this work, a numerical simulation experiment is introduced using two separate scale spaces. A scale-space model is constructed using blur-dependent image parameters that vary monotonically with the blur level. A polynomial model that fits blur-dependent parameters as a function of the blur level, with the undegraded image at the zeroth position is referred to as the first level scale space. Second level scale space is constructed with reference to one of the first level degraded image as the new reference. pixel-wise polynomial model representative of the second level scale space is then constructed to fit the spatially averaged distances at the successive blur levels. The estimation of error is calculated by extrapolating the second level polynomial to the reference point corresponding to the undegraded image. This error is referred to as deblurring error. Another type of error, fitting error is calculated from the mathematical formulation. From this work it is revealed that the deblurring errors and the fitting errors are large along the edges. Image sharpening can be performed only after minimizing these errors.