CS 474674 Image Processing Final Exam Review Fall

CS 474/674 – Image Processing Final Exam Review Fall 2019 - Prof. Bebis

Final Exam • When: 12/16/2019, 9: 50 am – 11: 50 am (same room) • What: comprehensive • Closed-books, closed-notes – True/False questions, 1 -2 proofs, and problems similar to the ones you have seen in class and the homework. – Study very well all the examples we did in class, quizzes, homework, and sample exams. • Bring a calculator (no smart phones!) • Be familiar with common equations (e. g. , FT, convolution, etc. ) - no need to memorize long/complicated formulas (e. g. , motion blur formula) – they will be provided to you if needed.

Final Exam Topics (comprehensive) 1. Midterm Exam material 2. Frequency Filtering 3. Image Restoration 4. Image Compression 5. Short-time Fourier Transform

Frequency Filtering • Low-pass, high-pass, and band-pass filters – Explain both in time and frequency domains • How is a filter specified in the frequency domain? – Specify in the time domain, then take FT (preserve symmetries!) – Specify directly in the frequency domain (sample correctly!) • What are the main steps of filtering in the frequency domain? • Why does low-pass filters cause blurring? – Explain both in time/frequency domains

Frequency Filtering (cont’d) • What is the “ringing” effect? – How can it be avoided? • Explain high emphasis filtering using frequency domain analysis. • What is homomorphic filtering? – What are the main assumptions? – When should one use it? – What are the main steps?

Image Restoration • What is the goal of image restoration? – How is it different from image enhancement techniques? • Degradation model under the assumptions of linearity and shift invariance – How do we model it? – Graduate Students Only: need to know the proof. • Noise models (probabilistic) – How could we estimate the model parameters? • Noise removal filters – mean, order statistics, adaptive, frequency domain

Image Restoration (cont’d) • Modeling the degradation function (e. g. , atmospheric turbulence, camera motion) – Graduate Students Only: need to know how to derive the degradation function in the case of motion blur. • Inverse filtering – How does it work? – Main assumptions? – Practical problems and ways to address them.

Image Restoration (cont’d) • Wiener filtering – – How does it work? Assumptions? How does it compare to inverse filtering? Practical problems and ways to address them. • Least Squares Constrained (LSC) filtering – – How does it work? Assumptions? How does it compares to Wiener filtering? What is its main advantage?

Image Compression • • What is the goal image compression? Lossless vs lossy compression Compression ratio, data redundancy, Data ≠ information Types of data redundancy – How can we deal with coding redundancy? (e. g. , variable length coding) – How can we deal with interpixel redundancy? (e. g. , mapping) – How can we deal with psychovisual redundancy? (e. g. , quantization)

Image Compression (cont’d) • How do we modeling the information generation process? – Probabilistically - explain • How do we measuring information? – Entropy - explain • Computing redundancy (i. e. , using entropy information). • Different order estimates of entropy and their importance. • Main components of image compression (i. e. , encoder, decoder). • Fidelity criteria (i. e. , subjective, objective)

Image Compression (cont’d) • Main components of encoders (i. e. , mapper, quantizer, symbol encoder) and decoders (i. e. , inverse steps) • Lossless compression – Huffman, Arithmetic, LZW, Run Length, Bit-plane • Lossy compression – Examples of mappers (i. e. , FT, DCT, WT) – What is the key property of a "good" mapper?

Image Compression (cont’d) • JPEG compression – What are the steps? – Understand each step and its purpose very well! • Different modes of JPEG – know what they are and what they do. – – Progressive spectral selection algorithm Progressive successive approximation algorithm Hybrid algorithm Hierarchical

Short-time Fourier Transform • Limitations of Fourier Transform – Lack of simultaneous time-frequency localization – explain. – Not useful for analyzing non-stationary signals – explain. – Not efficient in handling discontinuities – explain. • Short Time Fourier Transform (STFT) – What is the main idea? – How does it address the issue of simultaneous time-frequency localization? • Practical issues – Window function, window shape, window width. – Heisenberg (or Uncertainty) principle - explain