PSYCH 221 FINAL PROJECT

The film industry shares our problem as well. Their archived films have been degraded y vertical line scratches. The degradation process that creates these artifacts is difficult to mathematically model. Therefore, they treat the problem as a missing data problem.
To determine the missing data, two macro processes are performed: line scratch detection and line scratch removal. Since our problem has established which lines are missing, we can skip this step. To remove the line scratches there have been many different attempts, but the general process is quite consistent. An interpolation step is performed, and then a filter is applied. The interpolation step allows us to make a best guess estimate and the value of missing pixels. The filter step can be seen as two things: the removal of noise, and another interpolation step (if it is a linear filter and is only applied where the pixels are missing, it can be viewed as a linear interpolation).

For this experiment, we propose to use a bicubic interpolation for the first interpolation step. Then we used a variety of filters: Gaussian, Median, and Wiener. The Gaussian filter provides a common linear low-pass filter that helps smooth out data by removing detail and noise. Median filtering is a nonlinear operation often used in image processing to reduce noise but preserve edges. The filter operates by taking a block (5x5 in our case) around the current pixel, finding the median of the pixels, and then assigning the current pixel that value. Wiener filter is a low-pass adaptive filter based on statistics estimated from pixels surrounding the currently filtered pixel. It is quite common to use the wiener filter to improve an image that has been degraded by constant power additive noise. The use of a wide variety of filters is driven by the concept alluded to before that the ‘degradation process’ cannot be mathematically modeled. Therefore, we decided to try out all options and see which pair of interpolation and filters will work best over our test suite of images.