Presentation is loading. Please wait.

Presentation is loading. Please wait.

Lectures 10&11: Representation and description

Similar presentations


Presentation on theme: "Lectures 10&11: Representation and description"— Presentation transcript:

1 Lectures 10&11: Representation and description
Professor Heikki Kälviäinen Machine Vision and Pattern Recognition Laboratory (MVPR) Department of Information Technology Faculty of Technology Management Lappeenranta University of Technology (LUT) Prof. Heikki Kälviäinen CT50A6100 Machine Vision and Dig. Image Analysis

2 Task: Describe the region based on the chosen representation
Motivation Task: Describe the region based on the chosen representation Image acquisition => digital image Preprocessing => better image Segmentation => basic features Representation and description => advanced features Object recognition (classification, clustering) Prof. Heikki Kälviäinen CT50A6100 Machine Vision and Dig. Image Analysis

3 Content Representation schemes.
Chain codes, polygonal approximates, signatures, boundary segments, the skeleton of the region. Boundary descriptors. Simple descriptors (length, diameter, curvature), shape numbers, Fourier descriptors, moments. Regional descriptors. Simple descriptors (area, perimeter, compactness, principal axes), topological descriptors (holes, Euler number), texture, moments. Morphology: dilation, erosion, closing, opening. Relational descriptors. Structural relationships between ROIs. Prof. Heikki Kälviäinen CT50A6100 Machine Vision and Dig. Image Analysis

4 Representation schemes
Schemes that compact data into representations that are considerably more useful in the computation of descriptors. Schemes: Chain codes. Polygonal approximates. Signatures. Boundary segments. The skeleton of the region. Prof. Heikki Kälviäinen CT50A6100 Machine Vision and Dig. Image Analysis

5 4-directional chain and 8-directional chain codes.
Chain codes are used to represent a boundary by a connected sequence of straight-line segments of the specified length and direction. 4-directional chain and 8-directional chain codes. The chain code of the boundary depends on the starting point. Redefine the starting point so that the resulting sequence of numbers forms an integer of minimum magnitude. Normalization for rotation: the first difference of the chain code instead of the code itself. Prof. Heikki Kälviäinen CT50A6100 Machine Vision and Dig. Image Analysis

6 Polygonal approximation
A digital boundary can be approximated with arbitrary accuracy by a polygon. Minimum perimeter polygons. A polygon of minimum perimeter fitted to the object boundary enclosed by cells. Splitting techniques. Subdivision of a segment successively into two parts until a given criterion is satisfied. Prof. Heikki Kälviäinen CT50A6100 Machine Vision and Dig. Image Analysis

7 A signature is a 1-D functional representation of a boundary.
Signatures A signature is a 1-D functional representation of a boundary. To reduce the boundary representation to a 1-D function. Invariance: Translation. Rotation. Scaling. Prof. Heikki Kälviäinen CT50A6100 Machine Vision and Dig. Image Analysis

8 Decomposing a boundary into segments.
Boundary segments Decomposing a boundary into segments. The convex hull H of an arbitrary set S is the smallest convex set containing S. The set H-S is called the convex deficiency D of the set S. Prof. Heikki Kälviäinen CT50A6100 Machine Vision and Dig. Image Analysis

9 The skeleton of a region
To reduce the structural shape to a graph. The skeleton of the region by a thinning algorithm. Medial axis transformation (MAT). The MAT of a region R with border B is as follows: For each point p in R we find the closest neighbor in B. If p has more than one such neighbor it is said to belong to the medial axis (skeleton) of R. MAT depends on the choice of a distance measure (e.g., Euclidean distance) and also computationally expensive => many other thinning algorithms have been proposed. Prof. Heikki Kälviäinen CT50A6100 Machine Vision and Dig. Image Analysis

10 Boundary descriptors Simple descriptors.
Length of a contour: number of pixels along the contour. Diameter of a boundary B (D = distance measure): Diam (B) = max[D(p_i,p_j)]. Curvature: the rate of change of slope. Shape numbers: the first difference of the smallest magnitude. Fourier descriptors. Moments: the shape of boundary segments described by moments (mean value and variance of a 1-D function). Prof. Heikki Kälviäinen CT50A6100 Machine Vision and Dig. Image Analysis

11 Area, perimeter, compactness, principal axes. Topological descriptors.
Regional descriptors Simple descriptors. Area, perimeter, compactness, principal axes. Topological descriptors. Holes, connected components, Euler number. Texture. Moments. Prof. Heikki Kälviäinen CT50A6100 Machine Vision and Dig. Image Analysis

12 Simple descriptors of a region
Area = the number of pixels contained within its boundary. Perimeter = the length of its boundary. Compactness = Perimeter^2 / Area Elongatedness = Area/(2d)^2 d = erosion steps before the region totally disappears Eccentricity = ((μ_20 – μ_02)^2 + 4μ_11)/ μ_00 Principal axes: ratio = major axis / minor axis. Prof. Heikki Kälviäinen CT50A6100 Machine Vision and Dig. Image Analysis

13 Prof. Heikki Kälviäinen CT50A6100
Machine Vision and Dig. Image Analysis

14 Topological descriptors
Topology is the study of properties a figure that are unaffected by any deformation, as long as there is no tearing or joining of the figure (also called rubber-sheet distortions). Number of holes in a region H. Number of connected components C. Euler number E = C - H. Prof. Heikki Kälviäinen CT50A6100 Machine Vision and Dig. Image Analysis

15 Texture content of a region.
Texture descriptor provides measures of properties such as smoothness, coarseness, and regularity. Three approaches. Statistical approach: moments, co-occurrence matrix. Structural approach: texture primitives, rules S->aS. Spectral approach: Fourier spectrum. Prof. Heikki Kälviäinen CT50A6100 Machine Vision and Dig. Image Analysis

16 Co-occurrence matrix Original image (upper left) :
skin, leaves, brick, stone. Segmentation based on: Contrast (upper right). Homogeneity (lower left). Standard deviation (lower right). Prof. Heikki Kälviäinen CT50A6100 Machine Vision and Dig. Image Analysis

17 Texture Fourier signatures for skin (top) and for leaves (bottom):  (left), r (right)
Prof. Heikki Kälviäinen CT50A6100 Machine Vision and Dig. Image Analysis

18 Moments Moment of order (p+q) of f(x.y):
m_pq = ∫ ∫ x^p y^q f(x.y) dx dy where the integrals are (-∞,∞) Central moments for a digital image f(x.y): μ_pq = ∑ ∑ (x – x_m)^p (y- y_m)^q f(x,y) x y where x_m = m_10/m_00 and y_m = m_01/m_00 Prof. Heikki Kälviäinen CT50A6100 Machine Vision and Dig. Image Analysis

19 Morphological image processing
A tool for extracting image components that are useful in the presentation and description of region such as boundaries, skeletons, the convex hull. Also useful in pre- and post processing such as morphological filtering, thinning, and pruning. Dilation: expand the region using an structural element. Erosion: shrink the region using an structural element. Prof. Heikki Kälviäinen CT50A6100 Machine Vision and Dig. Image Analysis

20 Morphology: Dilation and erosion
Image A Structural element B Dilation Erosion Prof. Heikki Kälviäinen CT50A6100 Machine Vision and Dig. Image Analysis

21 Morphology: Dilation and erosion
Basic definations (A)_x = {c| c = a + x, for a є A} B^ = {x| x = -b, for b є B} Dilation A + B = {x| (B^)_x ∩ A ≠ Ø} = {x| [(B^)_x ∩ A] belongs to the subset of A} The dilation of A by B is the set of all x displacements such that B^ and A overlap by at least one nonzero element. Erosion A – B = {x| (B^)_x belongs to the subset of A} The erosion of A by B is the set of all points x such that B, translated by x, is contained in A. Prof. Heikki Kälviäinen CT50A6100 Machine Vision and Dig. Image Analysis

22 Morphology: Opening and closing
A ◦ B = (A – B ) + B Closing A • B = (A + B) – B To cope with distortions (e.g, to fill gaps) and noise (e.g., to remove isolated large noise areas). Prof. Heikki Kälviäinen CT50A6100 Machine Vision and Dig. Image Analysis

23 Morphology: Other operators
Hit or miss (points found in A with B_1 and in the complement of A with B_2). Boundary extraction. Connected components. Convex hull. Thinning. Thickening. Skeletons. Pruning (“sophisticated” thinning and skeletonizing). Prof. Heikki Kälviäinen CT50A6100 Machine Vision and Dig. Image Analysis

24 Relational descriptors
To organize components to exploit any structural relationships that may exist between them. Primitive elements. → a ↓ b Rewriting rules. S → aA, A → bS, and A → b Tree presentation. For example, the relationship “inside of” Prof. Heikki Kälviäinen CT50A6100 Machine Vision and Dig. Image Analysis

25 Examples of applications
Medical image processing. Diabetes and retinal image analysis using machine vision. Industrial machine vision. Paper and board printability tests by machine vision in the paper making and printing industry. Biometrics. Image-based biometric person authentication. Prof. Heikki Kälviäinen CT50A6100 Machine Vision and Dig. Image Analysis

26 Task: Describe the region based on the chosen representation
Summary Task: Describe the region based on the chosen representation Image acquisition => digital image Preprocessing => better image Segmentation => basic features Representation and description => advanced features Object recognition (classification, clustering) Prof. Heikki Kälviäinen CT50A6100 Machine Vision and Dig. Image Analysis


Download ppt "Lectures 10&11: Representation and description"

Similar presentations


Ads by Google