Color Image Indexing Using BTC Author: Guoping Qiu Source: IEEE - - PowerPoint PPT Presentation

Color Image Indexing Using BTC

Author: Guoping Qiu Source: IEEE Transaction on Image Processing,

• Vol. 12, NO. 1, pp. 93-101, 2003

Speaker: Tzu-Chuen Lu

Outline

Color Image Retrieval Color Correlogram (CC) Block Truncation Coding (BTC) BTC Color Co-Occurrence Matrix (BCCM) Block Pattern Histogram (BPH) Experiments and Results Conclusions

Color I mages Retrieval – Color Histogram

1 2 3 4 5 1 3 5 6 5 1 5 2 5 1 1 5 2 2 5 3 1 3 5 6 5 1 5 2

Color Correlogram (CC)

Image: I Image: I’

… Dis. 1 2 3 4 1 2 3 4 … 1 2 3 4 1 2 3 4 1 … 3 5 … 6 5 … 1 5 … 2 … 1 5 1 3 5 2

155 142 50 10 12 20 22 70 75 100 31 175 80 34 32 221 5 88 45 31 230 12 30 20 88

150 150 65 10 10 10 35 65 65 150 35 150 65 35 35 200 10 65 35 35 200 10 35 10 65

1 1 1

… Dis. 1 2 3 4 1 2 3 4 … 1 2 3 4 1 2 3 4 1 0 3 2 4 3 … 3 5 0 1 2 3 … 6 5 … 1 5 … 3 2 6 1 2 … 5 1 0 2 3 1 5 1 3 5 2

Block Truncation Coding (BTC)

Image: I B1 B2 B3 B4

1 2 6 1 3 1 5 1 4 1 . 5

Average Bitmap: I’

BTC for Color Image Coding

Bitmap: I’ 8* 16 = 128 (bits) Image: I 16 + (4* 2)* 8 = 80 (bits) Mean Values

BTC Color Co-Occurrence Matrix (BCCM)

0: 24 1: 228 0: 55 1: 234 0: 91 1: 210.5 0: 18 1: 112

1 2 … 228 … 255 1 2 … 24 … 255

1

10 20 … 100 … 200 10 3 2 … 3 … 5 20 1 … 1 … 1 … … … … … … … 200 2 1 … 3 … 3

For Example

10 20 30 40 10 5 3 20 2 1 30 2 1 6 40 2 1

I 1 I 2

10 20 30 40 10 2 3 20 1 1 1 1 30 6 8 1 1 40 2 3

1 1 | 1 | ... 3 1 | 3 | 2 5 1 | 2 5 | ce tan Dis + + − + + + + − + + + − =

Block Pattern Histogram (BPH)

1 0 1 1 1 1 1 1 1 0 1 1 0 0 1 1 1 0 1 1 0 1 0 1 0 0 0 0 1 1 0 1 1 0 1 1 1 1 1 1 1 0 1 1 0 0 1 1 1 0 1 1 0 1 0 1 0 0 0 0 1 1 0 1

Indx. 1 2 5 6 … 256 Pr.

1

Indx. 1 2 3 … 256 Pr. 5 2 1 2 3

Bitmap: I’

For Example

10 20 30 40 10 5 3 20 2 1 30 2 1 6 40 2 1

I 1 I 2

Indx. 1 2 3 … 256 Pr. 5 2 1 2 3

10 20 30 40 10 2 3 20 1 1 1 1 30 6 8 1 1 40 2 3

Indx. 1 2 3 … 256 Pr. 3 1 0 3

+ + + − + + + + − + + + − = ) 1 1 | 1 | ... 3 1 | 3 | 2 5 1 | 2 5 | ( ce tan Dis

) 30 1 | 30 | ... 1 2 1 | 1 2 | 3 5 1 | 3 5 | ( + + − + + + + − + + + −

Experiments and Results - I

• 720 texture images
• 120 texture classes
• 300* 200 pixels
• A block with 100* 100

pixels

Experiments and Results - I

Experiments and Results - II

20,000 color images in DB Data Query

Experiments and Results - II

96 query images

Conclusions

Using a well known image coding technique

BTC to retrieve images

Two image features derived directly from

an image

The new method achieves coding and

retrieval simultaneously