17-01-2012, 01:12 PM
Fast and Cheap Color Image Segmentation for Interactive Robots
[attachment=16163]
Introduction
An important first step in many color vision tasks is to classify
each pixel in an image into one of a discrete number
of color classes. The leading approaches to accomplishing
this task include linear color thresholding, nearest neighbor
classification, color space thresholdingand probabilistic
methods.
Description of the Approach Color Space Transformation
Our approach involves the use of thresholds in a three dimensional
color space. Several color spaces are in wide
use, including Hue Saturation Intensity (HSI), YUV and
Red Green Blue (RGB). The choice of color space for classification
depends on several factors includingwhich is provided
by the digitizinghardware and utility for the particular
application.
Thresholding
The thresholding method described here can be used with
general multidensional color spaces that have discrete component
color levels, but for the purposes of discussion the
YUV color space will be used as an example. In our approach,
each color class is specified as a set of six threshold
values: two for each dimension in the color space, after
the tranformation if one is being used. The mechanism
used for thresholding is an important efficiency consideration
because the thresholding operation must be repeated
for each color at each pixel in the image. One way to check
if a pixel is a member of a particular color class is to use a
set of comparisons similar to
