color texture segmentation using feature distributions

31
COLOR TEXTURE SEGMENTATION USING FEATURE DISTRIBUTIONS

Post on 22-Dec-2015

223 views

Category:

Documents


1 download

TRANSCRIPT

Page 1: COLOR TEXTURE SEGMENTATION USING FEATURE DISTRIBUTIONS

COLOR TEXTURE SEGMENTATION USING

FEATURE DISTRIBUTIONS

Page 2: COLOR TEXTURE SEGMENTATION USING FEATURE DISTRIBUTIONS

Hexagonal-HSV

Page 3: COLOR TEXTURE SEGMENTATION USING FEATURE DISTRIBUTIONS

Cellular decomposition

(S,H)={ (0.665, 49.1), (0.665, 169.1), (0.665, 229.1), (0.665,289.1), and (0.665, 349.1) }

Page 4: COLOR TEXTURE SEGMENTATION USING FEATURE DISTRIBUTIONS

Cellular decomposition

V component is divided into 3 parts

with three interval s (0, 0.45), (0.45, 0.75) and (0.75, 1).

Thus, the 7 (S, H) pairs and the 3 V values can be combined into 21 initial colors.

Page 5: COLOR TEXTURE SEGMENTATION USING FEATURE DISTRIBUTIONS

Cluster

* The Euclidean distances between each pixel and the 21 respective initial color centers are then computed.

* Each pixel can then be assigned to one of the 21 initial

color clusters whose center has the smallest color distance to the pixel.

* The cluster with the largest error is then chosen to perform further splitting

Page 6: COLOR TEXTURE SEGMENTATION USING FEATURE DISTRIBUTIONS

?

This process can be iterated level by level.

The repetitive splitting will be stopped when the splitting level reaches 4 or the number of used colors is more than 192.

Quantized image

Page 7: COLOR TEXTURE SEGMENTATION USING FEATURE DISTRIBUTIONS

Cluster

*the nearest pair of color clusters will be merged until the distance between any two colors are greater than 26.

*PSNR value of the quantizatzed image is larger than 30

峰值訊號雜訊比 (peak signal to noise ratio: PSNR)

dBMSE

PSNR )255

(log102

10

*merged until the number of used colors reaches 50

Page 8: COLOR TEXTURE SEGMENTATION USING FEATURE DISTRIBUTIONS

DEFINITION FEATURE

Color histogram : distribution of color in a texture

Local_edge_pattern histogram : distribution fo local edge patterns in a texture region

Page 9: COLOR TEXTURE SEGMENTATION USING FEATURE DISTRIBUTIONS

Color histogram hc

the color histogram hc for a texture region R using the following equation.

where n is the number of pixels with color label i, N is the total number of pixels of the texture region.

Page 10: COLOR TEXTURE SEGMENTATION USING FEATURE DISTRIBUTIONS

Local edge pattern descriptors

Page 11: COLOR TEXTURE SEGMENTATION USING FEATURE DISTRIBUTIONS

Local edge pattern (LEP)

he for a texture region R can be derived using the

following equation

where n is the number of pixels with LEP value i, N is the total number of pixels ofthe texture region.

Page 12: COLOR TEXTURE SEGMENTATION USING FEATURE DISTRIBUTIONS

Homogeneity Measure(color)

the more the Hc value approaches to 1

the more alike the color histograms hc and hc

Page 13: COLOR TEXTURE SEGMENTATION USING FEATURE DISTRIBUTIONS

Homogeneity Measure(texture)

the more the He value approaches to 1

the more alike the color histograms he and h e

Page 14: COLOR TEXTURE SEGMENTATION USING FEATURE DISTRIBUTIONS

Homogeneity Measure

Page 15: COLOR TEXTURE SEGMENTATION USING FEATURE DISTRIBUTIONS
Page 16: COLOR TEXTURE SEGMENTATION USING FEATURE DISTRIBUTIONS
Page 17: COLOR TEXTURE SEGMENTATION USING FEATURE DISTRIBUTIONS

Segmentation method

Page 18: COLOR TEXTURE SEGMENTATION USING FEATURE DISTRIBUTIONS

Hierarchical splitting

a block is split to four subblocks,six pairwise H distance as specified in Eq. (6) between color and LEP histograms of the four subblocks.

choose a small value

The threshold X was experimentally set to value 1.1.

Page 19: COLOR TEXTURE SEGMENTATION USING FEATURE DISTRIBUTIONS

Agglomerative merging

p is the number of pixels in the smaller of the two regions and H is the histogram distance measure as specified in Eq.(7)

respective color and LEP histograms are summed to the newimage region

Page 20: COLOR TEXTURE SEGMENTATION USING FEATURE DISTRIBUTIONS

Agglomerative merging

MImin is the smallest merger importance value of all preceding mergers, MImax is the largest merger importance value of all preceding

mergers MIcur is the merger importance for the current best merge.

Page 21: COLOR TEXTURE SEGMENTATION USING FEATURE DISTRIBUTIONS
Page 22: COLOR TEXTURE SEGMENTATION USING FEATURE DISTRIBUTIONS

Pixelwise classification

discrete disk with radius r and compute the corresponding color and LEP histograms over the disc

v is the number of pixels belonging to the ith region and in the 4-neighboors of the examined point.

Page 23: COLOR TEXTURE SEGMENTATION USING FEATURE DISTRIBUTIONS
Page 24: COLOR TEXTURE SEGMENTATION USING FEATURE DISTRIBUTIONS
Page 25: COLOR TEXTURE SEGMENTATION USING FEATURE DISTRIBUTIONS
Page 26: COLOR TEXTURE SEGMENTATION USING FEATURE DISTRIBUTIONS

Thank you~!!

Page 27: COLOR TEXTURE SEGMENTATION USING FEATURE DISTRIBUTIONS
Page 28: COLOR TEXTURE SEGMENTATION USING FEATURE DISTRIBUTIONS
Page 29: COLOR TEXTURE SEGMENTATION USING FEATURE DISTRIBUTIONS
Page 30: COLOR TEXTURE SEGMENTATION USING FEATURE DISTRIBUTIONS
Page 31: COLOR TEXTURE SEGMENTATION USING FEATURE DISTRIBUTIONS