Transcript

Semester Review10 Things You Have Learned

from EE465and

10 Things You Have NOT Learned from EE465

Apr. 28, 2011

10. MATLAB is an Evolving Language

• Just like any other programming language, MATLAB has its own evolutionary path – for better or for worse, remember it only serves as a tool

• Just like learning any natural or programming language, the only way to master it is through practice and emulation

9. MATLAB Programming Tips

• Know about MATLAB path setting and its limitations

• Avoid using loops/Vectorize your codes

• Functions run faster than scripts

• Load/save are faster than file I/O (including printing to the screen)

• Develop a good habit of putting comments

8. How JPEG Works

• JPEG=T+Q+C

• T: 8-by-8 Discrete Cosine Transform (admitting fast implementation)

• Q: quality factor (0-100) controls the tradeoff between compression ratio and image quality

• C: more/less frequent symbols, shorter/longer codewords

7. How Does Kmeans Work?

• Kmeans alternate between two steps– Update assignments: everyone vote for the

representatives (assigned to the nearest codeword)

– Update codewords: every precinct selects a new representative that best serves its voters (the center of the mass becomes the new representative)

6. What is Edge/Corner?

• There is no rigorous definition

• Canny edge detection might be the most popular/influential one in the engineering literature but it is not optimal

• From Harris’ corner detector to Lowe’s Scale-invariant feature transform (SIFT)– Sometimes perseverance pays off (Lowe

worked on his SIFT for many years)

5. Why Seeing is NOT Believing

• Theory– Context matters– Visual adaptation

• Applications– Data Hiding– Image forensics– Perceptive video coding (not as mature as MP3

for audio yet)

4. Hammer-Nail Match

• Tool sets: – FT, DCT, Hough/Radon transform …– median filtering, histogram equalization,

kmeans, …– edge/corner detection, SIFT …

• Problems sets:– Denoising: impulse vs. Gaussian noise – Coding vs. analysis

3. All Roads Lead to Rome

• Photo puzzle solver– Minimize the edge/corner count– Maximize the circle count (under the

assumption with circular image content)

• Image deblurring– Landweber vs. Lucy-Richardson

• Image denoising– Total-variation vs. Perona-Malik diffusion

2. Why Math is Useful?

• It turns art into science– Inverse filtering– Harris corner detection– Quantify uncertainty

• It serves as a universal language– Deterministic vs. Probabilistic– Analytical vs. geometric

1. Importance of Visual Information

• >70% information processed by human brain is visual in our everyday lives

• Higher quality, better life – acquisition, display, transmission and manipulation of images

• A vision view of brain – understanding vision could pave paths to understanding intelligence

• Visual reasoning (“a picture is worth a thousand words”)

Semester Review10 Things You Have Learned

from EE465and

10 Things You Have NOT Learned from EE465

Apr. 28, 2011

10. MATLAB speed up via MEX

• You can implement a MATLAB function by C or C++ and compile your .c/.cpp using MEX

• MEX compiler create a .dll (under windows) file that can be directly called by other MATLAB functions

• More information can be found at http://www.mathworks.com/support/tech-notes/1600/1605.html

9. MATLAB Profiling/Debugging

• Matlab profiling helps you locate where the computational bottleneck is

• Debugging in MATLAB has become more and more convenient (the interface is similar to that in Visual studio)

http://www.mathworks.com/help/techdoc/matlab_env/f9-17018.html

8. How JPEG2000 Works

• JPEG2000=T+Q+C

• T: Discrete Wavelet Transform (admitting fast implementation and progressive transmission)

• Q: rate-scalable

• C: replace Huffman coding by Arithmetic coding (covered by EE565)

7. How Does MRI Work?

6. How Does Biometrics Work?

• “You are your identity”

• Fingerprint: edge detection

• Iris: circle/pupil detection

• Face: SIFT matching

• Biometrics=sensor technology + image processing + pattern recognition

5. How Does Augmented Reality Work?

4. Why is Image Retrieval So Hard?

• Google

• Bing

• Yahoo

• TinEye

• Picsearch

3. Why is Color So Hard?

• Famous scientists who have worked on color: Newton, Maxwell, Schrödinger,

• Famous artists have their way of working with color: Vincent Van Gogh

• From dichomatic to trichomatic vision

2. Why is Math NOT Always Good?

• Mentally reproducible vs. experimentally reproducible– “It doesn't matter how beautiful your theory is,

it doesn't matter how smart you are. If it doesn't agree with experiment, it's wrong” - Richard Feynman

• If it is abused without physical intuition– Unfortunately it is difficult to detect

1. Everything is Connected• Image <> motion <> language

• Steve Jobs’ college life: machintosh inspired by what he learned from the caligraphy class

• Social network vs. neural network vs. immune network vs. Ising model


Top Related