Tony Li's CS180/Project 2

Gradient magnitude computation: measures the rate of change in a function by calculating the Euclidean norm of its gradient vector. It quantifies the steepness of the function at a given point. In image processing, we use the Euclidean norm of image convoluted with Dx and image convoluted with Dy to get gradient maginitude, which can be used for edge detection, since it reflects instantenous frequency changes.

Part 1.2 Comment: Gaussian kernel effectively make the two images convoleved with Dx and Dy more edgy. Originally convolution with Dx doesn't make the background building vertical edges noticable, but the gaussian filter makes them clearer. Facial expressions are easier to recoginize with filter + convolution. The final result has thicker lines of egdes and make the background objects recognizable.

Part 2.1 Comment: If we resharpen a blurred image, the resulting image can look a little bit shaper and edgy, but not clearer. If a sharp image is blurred and resharpened again, the resulting image can be only as good as the blurred image, but rarely able to be like the original image. The high frequency signals are already lost during the process, and this is hard to recreate the high frequency. This is consistent with expectations due to information loss.

Bells and Whistles: Skipped