1 / 28

Lecture 3: Region Based Vision

Lecture 3: Region Based Vision. Dr Carole Twining Thursday 18th March 1:00pm – 1:50pm. Segmenting an Image. Assigning labels to pixels (cat, ball, floor) Point processing: colour or grayscale values, thresholding Neighbourhood Processing: Regions of similar colours or textures

zander
Download Presentation

Lecture 3: Region Based Vision

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Lecture 3:Region Based Vision Dr Carole Twining Thursday 18th March 1:00pm – 1:50pm

  2. Segmenting an Image Assigning labels to pixels (cat, ball, floor) • Point processing: • colour or grayscale values, thresholding • Neighbourhood Processing: • Regions of similar colours or textures • Edge information (next lecture) • Prior information: (model-based vision) • I know what I expect a cat to look like

  3. Overview • Automatic threshold detection • Earlier, we did by inspection/guessing • Multi-Spectral segmentation • satellite and medical image data • Split and Merge • Hierarchical, region-based approach • Relaxation labelling • probabilistic, learning approach • Segmentation as optimisation

  4. Automatic Threshold Selection

  5. Automatic Thresholding: Optimal Segmentation Rule • Assume scene mixture of substances, each with normal/gaussian distribution of possible image values • Minimum error in probabilistic terms • But sum of gaussians not easy to find • Doesn’t always fit actual distribution Image Histogram

  6. T 4 3 Automatic Thresholding: Otsu’s Method • Extend to multiple classes

  7. T Automatic Thresholding: Max Entropy • Makes two sub-populations as peaky as possible

  8. cat floor combine Automatic Thresholding: Example ball

  9. Automatic Thresholding: Summary • Geometric shape of histogram (bumps, curves etc) • Algorithm or just by inspection • Statistics of sub-populations • Otsu & variance • Entropy methods • Model-based methods: • Mixture of gaussians • And so on. > 40 methods surveyed in literature • Fundamental limit on effectiveness: • Never give great result if distributions overlap • Whatever method, need further processing

  10. Multi-Spectral Segmentation

  11. f2 f1 Multi-spectral Segmentation • Multiple measurements at each pixel: • Satellite remote imaging, various wavebands • MR imaging, various imaging sequences • Colour (RGB channels, HSB etc) • Scattergram of pixels in vector space: • Can’t separate using single measurement • Can using multiple

  12. Spectral Bands Multi-Spectral Segmentation:Example Over-ripe Orange Scratched Orange Multispectral Image Segmentation by Energy Minimization for Fruit Quality Estimation: Martínez-Usó, Pla, and García-Sevilla,Pattern Recognition and Image Analysis , 2005

  13. Split and Merge

  14. Combine A D B C Split and Merge • Obvious approaches to segmentation: • Start from small regions and stitch them together • Start from large regions and split them • Start with large regions , split non-uniform regions • e.g. variance 2 > threshold • Merge similar adjacent regions • e.g. combined variance 2 < threshold • Region adjacency graph • housekeeping for adjacency as regions become irregular • regions are nodes, adjacency relations arcs • simple update rules during splitting and merging A & B

  15. Split Original Split Split Merge Merge Split and Merge

  16. Original Merge Split and Merge: Example Splitting

  17. Relaxation Labelling

  18. given that B is the case probability of A Mammals Pets ALL ALL ALL dog whale fish cat All Animal Species ( + ) ( + ) ( + ) x = x ( + ) ( + ) ( + ) ( + ) Aside: Conditional Probability • P(pet) = etc • P( pet | mammal) = • P( mammal | pet) = • Bayes Theorem: • P( pet | mammal )P(mammal) = P( mammal | pet )P(pet)

  19. Values from object pixels threshold Values from background Overlap: mistakes in labelling Relaxation Labelling: • Image histogram, object/background Context: Context to resolve ambiguity Label assignments

  20. Relaxation Labelling • Evidence for a label at a pixel: • Measurements at that pixel (e.g., pixel value) • Context for that pixel (i.e., what neighbours are doing) • Iterative approach, labelling evolves • Soft-assignment of labels: • Soft-assignment allows you to consider all possibilities • Let context act to find stable solution

  21. If not neighbours no effect Neighbours & same label support Neighbours & different label oppose all possible labels & how strong look at all other pixels degree of compatibility Relaxation Labelling • Compatibility: • Contextual support for label  at pixel i:

  22. Threshold labelling Noisy Image After iterating Relaxation Labelling: • Update soft labelling given context: • The more support, more likely the label • Iterate

  23. = 0.75 Trees Fields = 0.90 Initialisation Relaxation Labeling: • Value of  alters final result

  24. Segmentation as Optimisation

  25. Segmentation as Optimisation • Maximise probability of labelling given image: • Re-write by taking logs, minimise cost function: • How to find the appropriate form for the two terms. • How to find the optimum. label at i given value at i label at i given labels in neighbourhood of i label consistency label-data match

  26. label consistency label-data match E-step Model Parameters Image & Labelling M-step Segmentation as Optimisation • Exact form depends on type of data • Histogram gives: • Model of histogram (e.g., sum of gaussians, relaxation case) Learning approach: • Explicit training data (i.e., similar labelled images) • Unsupervised, from image itself (e.g., histogram model): Expectation/Maximization • Given labels, construct model • Given model, update labels • Repeat

  27. cost label values Segmentation as Optimisation label-data match term label consistency • General case: • High-dimensional search space, local minima • Analogy to statistical mechanics • crystalline solid finding minimum energy state • stochastic optimisation • simulated annealing • Search: • Downhill • Allow slight uphill

  28. Trees Fields = 0.90 Relaxation Optimisation Original Segmentation as Optimisation

More Related