This document outlines techniques for content-based image retrieval (CBIR) using color descriptors. It describes three color descriptor algorithms - columnar mean, average RGB method, and color moments. It evaluates these algorithms on a CBIR system and measures their performance in terms of precision, recall, and f-measure. The average RGB method achieved the highest average precision, recall, and f-measure. The document also discusses applying CBIR for retrieving images of marine invertebrates and compares different techniques for detecting features in images.
2. Outline
• Introduction
• Colourful Descriptors
– Columnar Mean
– Average RGB method
– Color Moments
• Results
• Comparison
• Categories for Retrieval
• Techniques for Detecting
• Working
• Comparison
• Inference
• Conclusion
• References 1
3. Introduction
• Explosive growth of image archive libraries
• Many CBIR methods proposed
• Works on the basis of similar images available
in the database
• Colour Descriptor: A fascinating feature
3
5. Columnar Mean
• Separate 3-D image colour planes in RGB planes
• Calculate row and column mean for each plane
• Find average mean to form a feature vector
• Compute Euclidean distance
• Retrieve Image from database on the basis of least distance vector
5
6. Average RGB Method
• Generate histogram of each plane
• Calculate average value of RGB
• Compute Euclidean Distance for image
similarity
• Retrieve images from database having smaller
distance to the query image
6
7. Color Moments Algorithm
• Convert RGB to HSV color
• For each plane, calculate
– Mean
– Standard Deviation
– Skewness
• Store feature vector value obtained through
nine moments
• Calculate distance
• Retrieve images having smallest vector 7
8. Results
• CBIR system deployed for every colour
descriptor algorithm
• Performance Measure
– Precision
– Recall
– f_measure
8
11. Categories for Retrieval
For the study of Marine Invertebrates:
• The taxonomic table containing taxa’s name
and nomenclature
• The geographical table that includes data of
museum catalogue collections or field books
• The table on ecology
• The table on bibliography
11
16. Conclusion
• Abundant of flora and fauna
• Proposed method is 98% precise
• Allows a query image
• Can be used by students for study purpose
• Performance calculated using Precision, Recall
and f_measure index
• Avg. RGB method- More significant
16
17. Future Scope
• Colourful Image Descriptors fused with
feature extraction methods like texture for
further investigating the accuracy and
efficiency.
17
18. References
[1] Kamlesh Kumar,Jian-Ping Li, Zain-ul-abiding,
Imran Khan, A Comparative Study
AmongColorful Image Descriptors for Content
Based Image Retrieval, IEEE, 2016.
[2]Mas Rina Mustaffa, Noris Mohd Norowi, and
Sim May Yee, Content-based Image Retrieval
System for Marine Invertebrates, IEEE, 2016.
12