Image Segmentation Image Analysis Object Recognition Image

Image Segmentation

Image Analysis: Object Recognition Image Segmentation INPUT IMAGE OBJECT IMAGE Image Segmentation: each object in the image is identified and isolated from the rest of the image

Image Analysis: Object Recognition Feature Extraction OBJECT IMAGE x 1 x 2 FEATURE VECTORS x … 3 xn Feature Extraction: measurements or “features” are computed on each object identified during the segmentation step

xn x 2 x 1 The feature vector for a given pixel consists of the corresponding pixels from each feature image; the feature vector for an object would be computed from pixels comprising the object, from each feature image.

Image Analysis: Object Recognition FEATURE VECTORS Classification OBJECT TYPE “WRENCH” Classification: each object is assigned to a class

Image Analysis: Object Recognition Image Segmentation INPUT IMAGE OBJECT IMAGE Feature Extraction FEATURE VECTOR Classification OBJECT TYPE “WRENCH”

Example: an automated fruit sorting system

Example: an automated fruit sorting system segmentation: identify the fruit objects the image is partitioned to isolate individual fruit objects

Example: an automated fruit sorting system segmentation: identify the fruit objects feature extraction: compute a size and color feature for each segmented region in the image size - diameter of each object color - red-to-green brightness ratio (redness measure)

Example: an automated fruit sorting system segmentation: identify the fruit objects feature extraction: compute a size and color feature for each segmented region in the image classification: partition the “fruit” objects in feature space

Automatic (unsupervised) image Segementation : difficult problem 1) attempt to control imaging conditions (industrial applications) 2) choose sensor which enhance objects of interest (infared imaging)

Two Types of Segmentation Algorithms: - Identify discontinuities between homogeneous regions - Identify similarity of pixel values within a region

Discontinuity based Segmentation Algorithms: Identify the boundaries between differing regions in the image. Two popular techniques use: - Spatial filters, gradients, edge linking - Identification of zero-crossings, thresholding

Discontinuity based Segmentation: detect points, lines and edges in an image

Discontinuity based Segmentation: detect points, lines and edges in an image -1 -1 8 -1 -1

Discontinuity based Segmentation: detect points, lines and edges in an image -1 -1 8 -1 -1 2 2 2 -1 -1 2 -1 -1 -1 2 -1 -1 -1 2

Discontinuity based Segmentation: detect points, lines and edges in an image -1 -1 8 -1 -1 2 2 2 -1 -1 2 -1 -1 0 1 -2 0 2 -1 0 1 -1 -1 2 -1 -1 -1 2 -1 -2 -1 0 0 0 1 2 1

Discontinuity based Segmentation: detect points, lines and edges in an image -1 -2 -1 0 0 0 1 2 1 -1 0 1 -2 0 2 -1 0 1 Gx Gy

Discontinuity based Segmentation: Gradient vector Gx Gy Edge Linking - used to create connected boundaries

Discontinuity based Segmentation: Gradient vector Gx Gy Edge Linking - used to create connected boundaries - similar points within a neighborhood are linked

Discontinuity based Segmentation: Gradient vector Gx Gy Edge Linking - used to create connected boundaries - similar points within a neighborhood are linked magnitude of gradient vector [ Gx 2 + Gy 2 ] 1 2

Discontinuity based Segmentation: Gradient vector Gx Gy Edge Linking - used to create connected boundaries - similar points within a neighborhood are linked magnitude of gradient vector [ Gx 2 + Gy 2 ] 1 2 approximated as | Gx | + | Gy |

Discontinuity based Segmentation: Gx Gy Gradient vector Edge Linking - used to create connected boundaries - similar points within a neighborhood are linked magnitude of gradient vector orientation of edges -1 ang(x, y) = tan ( Gy ) Gx

Discontinuity based Segmentation: Gradient vector Gx Gy Edge Linking - used to create connected boundaries - similar points within a neighborhood are linked magnitude of gradient vector orientation of edges

Discontinuity based Segmentation: Identify zero crossings

Discontinuity based Segmentation: Identify zero crossings 0 -1 4 -1 0 Laplacian Filter

Discontinuity based Segmentation: Identify zero crossings 0 0 -1 0 0 0 -1 -2 -1 0 -1 -2 16 -2 0 0 -1 -2 -1 0 0 0 -1 0 0 Laplacian Of a Gaussian

Discontinuity based Segmentation: Identify zero crossings Original image Lo. G

Discontinuity based Segmentation: Identify zero crossings Original image Thresholded Lo. G Outline of Thresholded Lo. G

Similarity based Segmentation: - Simple thresholding - Split and Merge - Recursive thresholding

Single Level Thresholding T[g] = 0, g < TH G - 1, TH # g

Single Level Thresholding T[g] = 0, g < TH G - 1, TH <= g

Single Level Thresholding

Single Level Thresholding T[g] = 0, g < TH G - 1, TH <= g

Multiple Level Thresholding T[g] = 0, g < TH 1 G - 1, TH 1 <= g <= TH 2 0, g > TH 2

Similarity based Segmentation: - Simple thresholding - Split and Merge - Recursive thresholding

Split and Merge 1) split region into four disjoint quadrants if P(Rj) = FALSE 2) merge any adjacent regions Rj and Rk if P(Rj URk) = TRUE 3) stop when no splitting or merging is possible

Split and Merge