1 / 29

Image Foresting Transform

Image Foresting Transform. for Image Segmentation. Presented by: Michael Fang Weilong Yang. A Few Things to Recall. Image Segmentation Finding homogeneous regions Graph-based Methods Treating images as graphs Image Foresting Transform Unification Efficiency Simplicity.

tricia
Download Presentation

Image Foresting Transform

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. Image Foresting Transform for Image Segmentation • Presented by: • Michael Fang • Weilong Yang

  2. A Few Things to Recall • Image Segmentation • Finding homogeneous regions • Graph-based Methods • Treating images as graphs • Image Foresting Transform • Unification • Efficiency • Simplicity

  3. Graph-Based Methods G ={V, E} V: graph nodes E: edges connection the nodes Pixels Pixel Similarity Segmentation = Graph Partition

  4. Directed Graphs A directed graph is a pair (I, A), where I is a set of nodes and A is a set of ordered pairs of nodes.

  5. Paths • A path is a sequence t1, t2, …, tk of distinct nodes in the graph, such that (ti, ti+1)  A for 1  i k – 1. • A path is trivial if k = 1; • Path    denotes the concatenation of two paths,  and , where  ends at t and  begins at t. • Path  =   s, t denotes theconcatenation of the longest prefix  of  and the last arc (s, t).

  6. Path Costs • A path-cost function is a mapping that assigns to each path  a cost (), in some ordered set  of cost values. • A function  is said to be monotonic-incremental (MI) when (t) = h(t), (  s, t) = ()  (s, t),where h(t) is a handicap cost value and  satisfies: x’  x  x’  (s, t)  x’  (s, t) and x  (s, t)  x,for x, x’   and (s, t)  A.

  7. Examples of MI Cost Functions • Additive cost function sum(t) = h(t), sum(  s, t) = sum() + w(s, t),where w(s, t) is a fixed non-negative arc weight. • Max-arc cost function max(t) = h(t), max(  s, t) = max{max(),w(s, t)}, where w(s, t) is a fixed arc weight.

  8. Predecessor Map and Spanning Forest • A predecessor map is a function P that assigns to each node t I either some other node in I, or a distinctive marker nil  I – in which case t is the root of the map. • A spanning forest is a predecessor map which takes every node to nil in a finite number of iterations (i.e., it contains no cycles).

  9. Paths of the Forest P • For any node t I, there is a path P*(t) which is obtained in backward by following the predecessor nodes along the path. P*(c) = a, b, c, where P(c) = b, P(b) = a, P(a) = nil P*(i) = i, where P(i) = nil

  10. Optimum-path Forest An optimum-path forest is a spanning forest P, where (P*(t)) is minimum for all nodes t I. Consider cost function sum in the example below.

  11. An Image as a Directed Graph • A grayscale image I is a pair (I, I), where I is a finite set of pixels (points in Z2) and I assigns to each pixel t I a value I(t) in some arbitrary value space. • An adjacency relation A is a binary relation between pixels of I, which is usually translation-invariant. • Once A has been fixed, image I can be interpreted as a directed graph, whose nodes are the image pixels in I and whose arcs are defined by A.

  12. Seed Pixels In some applications, we would like to use a predefined path-cost function  but constrain the search to paths that start in a given set SI of seed pixels. This constraint can be modeled by defining

  13. IFT algorithm for Image Segmentation 1. Path Cost 2. Four-Connected Adjacency

  14. IFT algorithm with FIFO policy(1) Initialization C(t) I t

  15. IFT algorithm with FIFO policy(2) ∝ ∝ ∝ ∝ 5 5 5 5 1 2 3 4 ∝ ∝ ∝ ∝ 5 5 5 5 5 6 7 8 ∝ ∝ ∝ ∝ 5 5 5 5 9 10 11 12 ∝ ∝ 0 0 0 5 5 0 0 0 13 14 15 16 13 0 0 16

  16. IFT algorithm with FIFO policy(3) Growing Process

  17. IFT algorithm with FIFO policy(4) ∝ ∝ ∝ ∝ 5 5 5 5 1 2 3 4 ∝ ∝ ∝ ∝ 5 5 5 5 5 6 7 8 5 ∝ ∝ ∝ 5 5 5 5 9 10 11 12 ∝ 0 5 0 0 5 5 0 13 14 15 16 5 5 0 5 5 0 16 9 14

  18. IFT algorithm with FIFO policy(4) ∝ ∝ ∝ ∝ 5 5 5 5 1 2 3 4 ∝ ∝ ∝ ∝ 5 5 5 5 5 6 7 8 5 ∝ ∝ 5 5 5 5 5 9 10 11 12 5 0 5 0 0 5 5 0 13 14 15 16 5 5 5 5 5 5 5 5 14 15 9 12

  19. IFT algorithm with FIFO policy(4) ∝ ∝ ∝ ∝ 5 5 5 5 1 2 3 4 5 ∝ ∝ ∝ 5 5 5 5 5 6 7 8 5 5 ∝ 5 5 5 5 5 9 10 11 12 5 0 5 0 0 5 5 0 13 14 15 16 5 5 5 5 5 5 5 5 5 5 12 15 10 14 5

  20. IFT algorithm with FIFO policy(4) ∝ ∝ ∝ ∝ 5 5 5 5 1 2 3 4 5 ∝ ∝ ∝ 5 5 5 5 5 6 7 8 5 5 ∝ 5 5 5 5 5 9 10 11 12 5 0 5 0 0 5 5 0 13 14 15 16 5 5 5 5 5 5 5 5 12 5 15 10

  21. IFT algorithm with FIFO policy(4) ∝ ∝ ∝ ∝ 5 5 5 5 1 2 3 4 5 ∝ ∝ 5 5 5 5 5 5 6 7 8 5 5 5 5 5 5 5 5 9 10 11 12 5 0 5 0 0 5 5 0 13 14 15 16 5 5 5 5 5 5 5 5 5 5 5 10 11 15 8

  22. IFT algorithm with FIFO policy(4) 5 5 5 5 5 5 5 5 1 2 3 4 5 5 5 5 5 5 5 5 5 6 7 8 5 5 5 5 5 5 5 5 9 10 11 12 5 0 5 0 0 5 5 0 13 14 15 16

  23. Another Example

  24. Framework of Image segmentation by IFT Input Image Gradient Image Seeds Labeling IFT

  25. Experiment Results (1)

  26. Experiment Results (2)

  27. Experiment Results (3)

  28. Experiment Results (4)

  29. Summary • Basic concept of the Image Foresting Transform • IFT for image segmentation • Experiment results

More Related