1 / 15

MPEG-7 DCD Based Relevance Feedback Using Merged Palette Histogram

MPEG-7 DCD Based Relevance Feedback Using Merged Palette Histogram. Ka-Man Wong and Lai-Man Po ISIMP 2004 Poly U, Hong Kong Department of Electronic Engineering City University of Hong Kong. A compact and effective descriptor Generated by GLA color quantization

gella
Download Presentation

MPEG-7 DCD Based Relevance Feedback Using Merged Palette Histogram

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. MPEG-7 DCD Based Relevance Feedback Using Merged Palette Histogram Ka-Man Wong and Lai-Man Po ISIMP 2004 Poly U, Hong Kong Department of Electronic Engineering City University of Hong Kong

  2. A compact and effective descriptor Generated by GLA color quantization Maximum of 8 colors in storage Each color have a minimum distance (Td) of 15 in CIELuv color space MPEG-7 Dominant Color Descriptor

  3. Commonly used RF techniques and limitations • Feature weighting relevance feedback technique • Assumes a fixed feature space (histograms) • Taking liner combinations on matching histogram bins. • Simple approach: Histogram averaging ( + ) / 2 =

  4. H1 H’ H2 Commonly used RF technique and limitations • But DCDs of images might have different set of colors, similar images might not have any exactly matched colors. • The number of colors in updated query may greatly exceed the limit of the number of colors defined by MPEG-7 • Similar colors are separated. By definition of DCD, similar colors should be grouped together.

  5. Limitation of feature weighting relevance feedbacktechnique • The Merged Palette Histogram Relevance Feedback • The updated query contains common colors among selected images • Represent the selected images efficiently

  6. Proposed Merged Palette Histogram for Relevance Feedback • Merged Palette Histogram Relevance Feedback (MPH-RF) process - initialize • Obtain all DCD from selected images

  7. + + = Proposed Merged Palette Histogram for Relevance Feedback • Merged Palette Histogram Relevance Feedback (MPH-RF) process - 1 • Link all DCD together 8 colors 6 colors 20 colors 6 colors

  8. Proposed Merged Palette Histogram for Relevance Feedback • Merged Palette Histogram Relevance Feedback (MPH-RF) process - 2 • Palette Merging • Find the closest pair of colorsbased on Euclidian distance in CIELuv • If minimum distance smaller than Td merge the color pair and sum up the percentages of merged colors • Iterate until minimum distance > Td 9 colors 20 colors

  9. Proposed Merged Palette Histogram for Relevance Feedback • Merged Palette Histogram Relevance Feedback (MPH-RF) process - 3 • Approximation • Cut the least significant colors if number of colors >8 9 colors 8 colors

  10. Approximated MPH Updated QueryHistogram Sum =1 Proposed Merged Palette Histogram for Relevance Feedback • Merged Palette Histogram Relevance Feedback (MPH-RF) process - 4 • Re-normalization • Adjust the histogram sum into 1 • An updated query is generated

  11. Experimental Results of MPH-RF • Experiment Methodology • ANMRR • Image Database • 5466 Image from MPEG-7 common color dataset (CCD) • Pre-defined queries and ground truth set • Relevance Feedback • Ground truth images are selected as relevant images

  12. Latest experimental results • MPH-RF gives improvement on both similarity measure methods • Combination of MPHSM and MPH-RF gives a significant improvement • Three iterations of relevance feedback give a significant result *ANMRR – smaller means better

  13. Query image Ground truth images First RF retrieval, 6 of 8 ground truths hit, NMRR=0.2782 Initial retrieval, 4 of 8 ground truths hit, NMRR=0.5 Second RF retrieval, 7 of 8 ground truths hit, NMRR=0.1541 Experimental results • Visual results – Query #50 from MPEG-7 CCD, MPHSM • Visit http://www.ee.cityu.edu.hk/~mirror for more results

  14. Experimental results • Visual results – Query #24 from MPEG-7 CCD, QHDM • Visit http://www.ee.cityu.edu.hk/~mirror for more results Query image Ground truth images First RF retrieval, 6 of 12 ground truths hit, NMRR=0.3738 Initial retrieval, 5 of 12 ground truths hit, NMRR=0.5125 Second RF retrieval, 9 of 12 ground truths hit, NMRR=0.1963

  15. Conclusions on Merged Palette Histogram Relevance Feedback • A new MPH-RF for MPEG-7 DCD is proposed • MPH-RF generates a new DCD query using palette merging technique • Represents the selected relevant images naturally and effectively • Experiment result also found that proposed method improve DCD-QHDM by 0.0717 and MPHSM by 0.0637 using MPEG-7 Common Color Dataset • The proposed method also provide better perceptually relevant image retrieval.

More Related