Corner Detection & Color Segmentation

Corner Detection & Color Segmentation

CSE350/450-0119 Sep 03

Administration

• Clarifications to Homework 1 • Questions?

Class Objectives

• Linear Algebra Review• Review how corners can be extracted

from computer images• Review how color is represented and

can be segmented in a computer image

Supporting References

• “A Tutorial on Linear Algebra” by Professor C. T. Abdallah, University of New Mexico

• Edge & Corner Detection: Introductory Techniques for 3-D Computer Vision, Trucco & Verri, 1998

• CVOnline “Color Image Processing” Lecture Notes• Poynton's Color FAQ

Edge Detection ReviewINPUT IMAGE

1) NoiseSmoothing

EDGE IMAGE

2) EdgeEnhancement

Horizontal [-1 0 1]

Vertical [-1 0 1]T

),( yxI

xyxI

),(

yyxI

),(

21

22 ),(),(),(

yyxI

xyxIyxI

“GRADIENT” IMAGE

3)Threshold

16/121242121

Linear Algebra Review

Corner Detection Motivation

• Corners correspond to point in the both the world and image spaces

• Tracking multiple point across consecutive images allows us to estimate the relative rotation and translation of the camera– Hartley’s 8-point algorithm

• Since the camera moves with our robot, we can infer robot motion “simply” by tracking eight or more corners

Corner Detection AlgorithmTrucco & Verri, 1998

61605319185855531513555550131310101011111012121110

yyxII

xyxII yx

),(,),(

1. Compute the image gradients

2. Define a neighborhood size as an area of interest around each pixel

3x3 neighborhood

3. For each image pixel (i,j), construct the following matrix from it and its neighborhood values

e.g.

Corner Detection Algorithm (cont’d)

61605319185855531513555550131310101011111012121110

xI

2

2

),(yyx

yxxji III

IIIC

22222

2222)3,3(

5553155550

13101011]1,1[

C

3. For each matrix C(i,j), determine the 2 eigenvalues λ(i.j)= [λ1, λ2].4. Construct Λ-image where Λ(i,j)=min(λ(i.j)).5. Threshold Λ-image. Anything greater than threshold is a corner.

Corner Detection Algorithm (cont’d)

ISSUE: The corners obtained will be a function of the threshold !

Corner Detection Sample ResultsThreshold=25,000 Threshold=10,000

Threshold=5,000

Color Segmentation Motivation

• Computationally inexpensive (relative to other features)

• “Contrived” colors are easy to track • Combines with other features for robust

tracking

What is Color?• Color is the perception of light in the visible

region of the spectrum• Wavelengths between 400nm - 700nm• Imagers

– Retina (humans)– CCD/CMOS (cameras)

RGB Color Space• Motivated by human visual system

– 3 color receptor cells (rods) in the retina with different spectral response curves• Used in color monitors and most video cameras

YCbCr (YUV/YIQ) Color Space

“Greyscale”Y= 0.30*R+0.59*G+0.11*B

BGR

VUY

081.0419.0500.0500.0331.0169.0114.0587.0299.0

• Separates luma (“brightness”) from the chroma (“color”) channels: Y = 0.30*R+0.59*G+0.11*B, Cb = B-Y, Cr=R-Y

• YUV/YIQ are similar variants based upon NTSC/PAL television signals

Defining Colors in an RGB Image

Red Green Blue

How do we represent a “single” color?

Sample set for orange hat

Simple RGB Color Segmentation

)1.1,5.254( )8.14,6.103( )07.6,1.45(

256),(251 yxIR 135),(73 yxIG 58),(32 yxIB

& &

Red Green Blue

SegmentedColor Image

Color Tracking Demo