Content-Based Image Retrieval (CBIR) Student: Mihaela David Professor: Michael Eckmann Most of the database images in this presentation are from the Annotated Image Database from Linda Shapiro's group at the University of Washington. This database can be found online here:
Project Description Searching and/or browsing a database of digital images based on the content of the image itself rather than on information about the image Searching and/or browsing a database of digital images based on the content of the image itself rather than on information about the image User searches database by providing a query image User searches database by providing a query image
Project Goals Develop a software system that can be used to research and compare CBIR techniques Develop a software system that can be used to research and compare CBIR techniques Implement and experiment with various techniques Implement and experiment with various techniques Evaluate techniques Evaluate techniques
Timeline Phase 1 – Initial decisions Phase 1 – Initial decisions Phase 2 – Software design / implementation Phase 2 – Software design / implementation Phase 3 – Experimentation with CBIR techniques Phase 3 – Experimentation with CBIR techniques Phase 4 – Graphical user interface Phase 4 – Graphical user interface
Phase 1 Matlab the most appropriate programming language for the project Matlab the most appropriate programming language for the project – Image Processing Toolbox – Graphical User Interface – Easy to code – Runs fast most cases – Easy to learn new language features
Phase 2 Implementing the software system Implementing the software system – Creates listing of image files in a directory and its subdirectories – Runs technique for all images in database – Compares query image to all the images in the database – Displays the image results sorted best match to worst match
Phase 3 Using the software to experiment with and evaluate a few approaches Using the software to experiment with and evaluate a few approaches Color Histograms Color Histograms – Peak Detection in a histogram – Texture
Phase 4 – Designing and implementing the graphical user interface multiple techniques multiple techniques query image selection query image selection browsing results browsing results
The interface
The interface (cont.)
Histograms – intensities – counts of pixels – position ignored
Color Histograms & Peaks Detection – HUE-SAT histogram – Peaks: dominant colors in image – Compare images by matching peaks
Peaks in HUE-SAT histogram
Pro's and Con's Speed of execution (+) Speed of execution (+) Compact representation (+) Compact representation (+) Intersection (+) Intersection (+) Independent of scale (+/-) Independent of scale (+/-) Ignores position (+/-) Ignores position (+/-)
Benefits To student: To student: – Programming experience – Matlab – Computer vision To professor: To professor: – Software system – Experimentation with CBIR techniques