Digital Image Processing

Catalog description

The first half of the course contains a detailed study of the mathematical tools required for understanding and implementing specific digital image processing algorithms such as an overview of the human visual system, Cartesian-separable vs. isotropic filters, fast approximation of Gaussian filters, a comprehensive review of 2-D digital spatial filters (LP, HP, sharpening, edge detection), the integral image, 2-D sampling strategies (e.g., Cartesian, Hexagonal, or general grid), fundamentals of image resizing (bilinear, bicubic, Lanczos, etc.), geometric transforms and image warping, and detailed coverage of 2-D discrete Fourier transform. The second half of the course focuses on specific digital image processing algorithms including contrast enhancement, noise reduction, sharpening, deblurring and segmentation. Some specific techniques for contrast enhancement are linear and nonlinear look-up tables, histogram equalization and modification, and contrast-limited adaptive HE (CLAHE). Algorithms for linear and nonlinear noise reduction include selective averaging, the sigma filter, the K-NN filter, bi-lateral filtering, median filtering, and deep networks. Sharpening techniques include nonadaptive and adaptive unsharp masking and relaxation of the boosting parameter. Deblurring techniques include the inverse filter and the Wiener filter. Finally, segmentation algorithms include various edge detection masks, the Otsu algorithm and adaptive thresholding. This course relies heavily on the knowledge of an undergraduate EE course in linear systems such as shift-invariant linear systems, impulse response, continuous and discrete Fourier transforms, the sampling theorem and the convolution operation. Additionally, EEEE678 serves as a good background or can be taken simultaneously.