CS663 - Digital Image Processing

Instructor: Ajit Rajwade
Office: SIA-218, KReSIT Building
Email:
Lecture Venue: LC 101
Lecture Timings: Slot 8, Monday and Thursday 2:00 to 3:30 pm
Instructor Office Hours (in room SIA-218): Tuesday and Friday 5:00 to 6:00 pm, or by appointment
Teaching Assistants: Sougata Sinha, Souvik Sinha Deb, Prerana Singhal, Alankar Kotwal, Thyagarajan Radhakrishnan, Raviteja Chalasani


Textbooks and Resources



Lecture Schedule:

(Continuing from the previous set of lectures taken by Prof. Suyash Awate)
Previous offerings of this course:

Date

Content of the Lecture

Assignments/Readings/Notes

14/09 (Mon)
  • Face recognition: intro; Principal components analysis for face recognition (eigenfaces): intro, concept of covariance matrix, description of algorithm and its computational complexity; a faster algorithm for PCA on a small (N) number of large-sized images (N << d case).
21/09 (Mon)
  • PCA: main principles, and derivation for k = 1 directions, and for k > 1 directions; choice of k in face recognition; concept of Lagrange multipliers
24/09 (Thurs)
  • Derivation sketch for k > 1 directions, concept of person/pose specific eigenspaces, a word about 3D face recognition; Concept of SVD (Singular Value Decomposition): reduced form, formula using outer products, applications to image compression; Eckart Young theorem
28/09 (Thurs)
  • SVD: properties (determinant, frobenius norm, rank, inverse and pseudo-inverse), implementing eigenfaces using SVD, overview of some other applications of SVD
    Image restoration: problem statement, simplifying assumptions; Models of blur: defocus and motion blur
1/10 (Thurs)
  • Derivation of Motion blur kernel under in-plane translation; Image restoration: inverse filter and problems with the inverse filter; spread-spectrum filters: coded aperture and flutter-shutter camera; Introduction to the Wiener filter
5/10 (Mon)
  • Introduction to the Wiener filter: principles on which it based, criterion that it optimizes
  • Derivation of Wiener filter, different variants of its formula
  • Interactive Wiener filter
  • Regularized deblurring filter (penalizing the derivatives)
  • Introduction to PCA for denoising
8/10 (Thurs)
12/10 (Mon)
  • Visible spectrum
  • Color image perception: the theory of human perception based on the three types of cones
  • RGB color model, CMY(K) color models; additive and subtractive color mixing; related optical illusion
  • HSI color model, the concept of hue, saturation and intensity, the illumination invariant property of hue
15/10 (Thurs)
  • HSI color model, the concept of hue, saturation and intensity, the illumination invariant property of hue
  • Advantages and disadvantages of hue
  • Concept of chromaticity vector
  • Color image processing: color image histogram equalization, color image bilateral filtering, concept of edge in a color image as an objective function using directional derivatives
  • PCA of RGB values: reiterating the concept that PCA is a decorrelating transform (to be continued later while explaining the YCbCr color model)
26/10 (Mon)
  • PCA of RGB values: reiterating the concept that PCA is a decorrelating transform
  • YCbCr and YUV color spaces
  • Hyperspectral image: concept, applications, visualization, PCA on hyperspectral image values
  • Concept of color filter arrays, Bayer filter, demosaicing; demosaicing algorithm using bilateral filter
  • Concept of compressed sensing and its relation to mosaicing and demosaicing
29/10 (Thurs)
  • Lossless and lossy compression, importance of lossy compression for images/video
  • Introduction to steps of the JPEG standard, concept of quality factor
  • Discrete cosine transform: definition, properties, advantages over DFT
  • Slides
  • From the book by Gonzalez: sections 8.2.1, 8.2.8 (skip portion on Walsh Hadamard Transform)
  • Section 5.6 from the book by Anil K Jain - for material pertaining to DCT, also see definition of Markov processes in section (2.9) (equations 2.67 and 2.68)
31/10 (Sat) (extra lecture)
  • Discrete cosine transform: definition, properties, advantages over DFT
  • DCT and its relationship with PCA for a stationary first order Markov process with $\rho$ close to 1
  • Quantization step in JPEG
  • Huffman encoding and decoding
2/11 (Mon)
  • Proof of relation between DCT and DFT
  • Huffman encoding and run length encoding in JPEG
  • JPEG compression for color images - in the YCbCr color space
  • Modes of JPEG compression
  • Introduction to video compression, concept of predictive encoding for video
5/11 (Thurs)
  • Introduction to video compression, concept of predictive encoding for video
  • First order differential encoding for video, motion compensated residuals
  • Concept of P,B,I frames in MPEG
  • Architecture of MPEG encoder and decoder
  • Introduction to the concept of compressed sensing (not on exam)