Transcript 2. Color Space

Performance Analysis of Color
Components
in Histogram-Based Image Retrieval
Presented by
Tienwei Tsai
Department of Information Management
Chihlee Institute of Technology,
2006/6/26
Outline
1. Introduction
2. Color Space
3. Color Histogram
4. Distance Measurement
5. Experimental Results
6. Conclusions
1. Introduction
• Two approaches for image retrieval:
– query-by-text (QBT): annotation-based image
retrieval (ABIR)
– query-by-example (QBE): content-based image
retrieval (CBIR)
• Drawbacks of ABIR:
– time-consuming and therefore costly.
– human annotation is subjective.
– some images could not be annotated because it is
difficult to describe their content with words.
• Standard CBIR techniques can find the images
exactly matching the user query only.
• In QBE, the retrieval of images has been done via
the similarity between the query image and all
candidates on the image database.
– Euclidean distance
• CBIR is a technology to search for similar images to
a query based only on the image pixel
representation.
– It is necessary to devise a means of describing the
location of each pixel and its intensity.
• How to choose a suitable color space is a critical
problem in image retrieval.
– In this paper, we investigate the appropriateness of the
three well-known color spaces, i.e., RGB, YUV, and HSV,
for CBIR.
• In our demo system,
– each image is first transformed from the standard RGB
color space into the YUV (or HSV) space.
– then, the histogram for each component (e.g., luminance Y,
blue chrominance U, and red chrominance V) of the image
is obtained, which can be served as the color feature of
the image.
• In the image database establishing phase
– the features of each image are stored;
• In the image retrieving phase
– the system compares the features of the query image with
those of the images in the database, using the Euclidean
distance metric and find out good matches.
2. Color Space
• A color space is a model for representing
color in terms of intensity values.
• There exist many models:
– RGB (Red, Green, and Blue),
– CMYK (Cyan, Magenta, Yellow, and Black
Key),
– YUV (Luminance and Chroma channels), and
– HSV (Hue, Saturation, and Value ), etc.
2.1 RGB Color Space
•
For a gray-level digital image
–
•
It can be defined as a function of two variables, f(x,
y), where x and y are spatial coordinates, and the
amplitude f at a given pair of coordinates is called
the intensity of the image at that point.
For a color image
–
each pixel (x, y) consists of three components: R(x,
y), G(x, y), and B(x, y), each of which corresponds
to the intensity of the red, green, and blue color in
the pixel, respectively.
2.2 YUV Color Space
•
Originally used for PAL (European "standard")
analog video
–
–
•
The Y primary was specifically designed to follow
the luminous efficiency function of human eyes.
Chrominance is the difference between a color and
a reference white at the same luminance.
The following equations are used to convert
from RGB to YUV color space:
Y  0.299 R  0.587G  0.114 B,
U  0.492( B  Y ),
V  0.877( R  Y ).
2.3 HSV Color Space
•
The HSV stands for the Hue, Saturation, and Value
based on the artists (Tint, Shade, and Tone).
–
–
–
•
As hue varies from 0 to 1.0, the corresponding colors vary
from red, through yellow, green, cyan, blue, and magenta,
back to red.
As saturation varies from 0 to 1.0, the corresponding colors
(hues) vary from unsaturated (shades of gray) to fully
saturated (no white component).
As value, or brightness, varies from 0 to 1.0, the
corresponding colors become increasingly brighter.
The conversion formula from RGB to HSV color space
is as follows:

 [( R  G)  ( R  B)] / 2 

H  cos 

2

 ( R  G)  ( R  B)(G  B) 

1
S  1  3  [min( R, G, B)] /( R  G  B),
V  ( R  G  B) / 3.
3 Color Histogram
•
The color histogram for an image is constructed by
discretizing (or quantizing) the colors within the image
and counting the number of pixels of each color.
More formally, it is defined as
•
h X ,Y , Z ( x, y, z )  N  Pr ob( X  x, Y  y, Z  z ),
•
For gray-scale images these are 2-D vectors.
–
•
As for color images
–
•
One dimension gives the value of the gray-level and the other
the count of pixels at the gray-level.
each color channel can be regarded as gray-scale images.
More generally, we can set the number of bins in the
color histograms to obtain the feature vector of desired
size.
•
To obtain the color feature of an image, some color components
can easily be analyzed according their histogram type.
–
•
One can realize the brightness of an image by checking the
histogram type of the color components such as G (for RGB), Y (for
YUV), and V (for HSV);
One can realize the dominant color of an image according to the
following heuristic rules:
1.
2.
3.
4.
5.
6.
7.
if the U component is high key, then the dominant color is blue;
if the U component is low key, then the dominant color is yellow;
if the V component of YUV space is high key, then the dominant color
is red;
if the V component of YUV space is low key, then the dominant color
is green;
if the H component is low key, then the dominant color is red;
if the H component is midtone, then the dominant color is green;
if the H component is high key, then the dominant color is blue, and
so on.
4. Distance Measurement
• To decide which image in the image
database is the most similar one with the
query image, we have to define a measure
to indicate the degree of similarity.
• Therefore, the distance (or dissimilarity)
between a feature vector Fm of the query
image and that of an image in the
database is based on the distance function.
• In our approach, the distance between two
vectors is calculated on the basis of the sum of
squared differences (SSD).
• Assume that and represent the mth feature of
the query image Q and an image X in the
database, respectively; each feature comes from
the color histograms.
• Then, the distance is
K 1
d m ( Fqm , Fxm )   (Fqm [i ]  Fxm [i ]) 2 ,
i 0
5. Experimental Results
• We evaluated performance on a test image
database, which was downloaded from the
WBIIS.
• It is a general-purpose database including 1,000
color images. The images are mostly
photographic and have various contents, such
as natural scenes, animals, insects, building,
people, and so on.
• In the experiment, an image retrieval demo
system is built to test the three color spaces
and their components .
Figure 2. The GUI of our demo system.
•
To compare the three color spaces and their components in a
quantitative manner, five classes of query images, referring to
–
•
To assess the ground-truth relevance score to each image for
each benchmark query, each target image in the collection is
assigned a relevance score as follows:
–
–
•
•
white owl (5 images), pumpkins (4 images), red apples (3 images),
green apples (3 images), and deer (9 images), are served as the
benchmark queries.
1 if it belonged to the same class as the query image,
0 otherwise.
The process was repeated for all the relevant images and an
overall average retrieval effectiveness was computed for each of
the color component and each of the query example.
The overall average relevance score in top n was computed by
averaging the individual values in each top n. The bin number
used for each color component is 5.
•
We can conclude that:
– The HSV and RGB color spaces perform
well in fine matching; however, the YUV
color space performs better in coarse
classification.
– Even though the B and U components are
less sensitive to human vision, they are
quite important in terms of the effectiveness
of the histogram-based retrieval.
– The V component performs well in coarse
classification.
6. Conclusions
•
•
•
•
•
In this paper, we investigate the appropriateness of the
three well-known color spaces, i.e., RGB, YUV, and
HSV, for histogram-based color image retrieval.
An image retrieval demo system is built to make it easy
to test the retrieval performance.
We have performed experiments on a database with
1000 images, using the Euclidean distance metric.
The results show that the HSV and RGB color spaces
perform well in fine matching; however, the YUV color
space performs better in coarse classification.
Moreover, even though the B and U components are
less sensitive to human vision, they are quite important
for the histogram-based retrieval.
Thank You !!!