A. Vadivel, M. Mohan, Shamik Sural and

1
SEGMENTATION USING SATURATION THRESHOLDING AND
ITS APPLICATION IN CONTENT-BASED RETRIEVAL OF
IMAGES

• A. Vadivel, M. Mohan, Shamik Sural and
• A. K. Majumdar
• INDIAN INSTITUTE OF TECHNOLOGY KHARAGPUR, INDIA.
• shamik_at_sit.iitkgp.ernet.in

2
ANALYSIS OF THE HSV COLOR SPACE

• Three dimensional representation of the HSV color
  space
• Central vertical axis represents intensity, I
• Hue, H, is an angle in the range 0,2p relative
  to the red axis.
• Saturation, S, is the depth or purity of color

3
VISUAL PERCEPTION IN THE HSV COLOR SPACE
(a)
(b)
Variation of color perception with (a) saturation
(Decreasing from 1 to 0 right to left) for a
fixed value of intensity and (b) intensity
(Decreasing from 255 to 0 right to left) for a
fixed value of saturation.
4
SATURATION THRESHOLDING

• Use Saturation of a Pixel to Determine
• Hue or Intensity is more pertinent to Human
  Visual Perception
• Ignore Actual Value Saturation
• Image pixels from a distribution of colors
• gray color and
• true color
• Threshold to determine a true color and a gray
  color pixel
• thsat(V) 1.0 (0.8V/255)

5
IMAGE REPRESENTATION

• Each image is represented as a collection of
  pixel features I?(pos, tg, val)
• val ?0,2p if tg takes a value of t and
• val ?0,255 if tg takes a value of g.
• Approximation occurs as shown below

(a) (b)
(c) (d)   (a) Original
Image (b) HSV Approximation (c) RGB approximation
with all low order bits set to 0 and (d) RGB
approximation with all low order bits set to 1.
6
IMAGE SEGMENTATION

• Pixel Grouping by K-means Clustering Algorithm
• I ?O1O2O3.On
• Oi?(tg, val, pos)
• Each partition represents either a true color
  value or a gray color value
• Consists of the positions of all the image pixels
  that have colors close to val.
• Post Processing
• Connected Component Analysis
• Filtering
• Merging

7
DIFFERENT STAGES OF IMAGE SEGMENTATION
(a) (b) (c)
(d)
(a) Original image (b) Image after clustering (c)
Image after connected component analysis and (d)
Final segmented Image
8
IMAGE SEGMENTATION RESULTS
Natural Scene Images
9
IMAGE SEGMENTATION RESULTS CONTD.
Miscellaneous Images
10
WEB BASED IMAGE RETRIEVAL APPLICATION

• Query Specification
• By Example Image
• Display of Result Set
• Nearest Neighbor Retrieval
• External Image Upload
• Utility to Upload External Image Files

11
PERFORMANCE MEASURE
Recall
Precision

• Relevant Set is not known for Large Uncontrolled
  Data Sets

Perceived Precision
12
RESULTS
Precision vs. recall on a controlled database of
2,015 images
Perceived precision variation on a large
un-controlled database of 28,168 images
13
CONCLUSIONS AND FUTURE WORK

• Effective way to capture Pixel Color Information
  using the HSV Color Space
• Relative importance of Hue and Intensity
  determined based on Saturation
• Pixel represented as True Color or Gray Color
• Clustering of Image Pixels for Segmentation
• Post-processing
• Web-based Image Retrieval System
• Perceived Precision for Testing on Large Data
  Sets
• More Experiments and Comparative Studies being
  Carried out
• New Distance Metrics

14
Thank You