Content-based retrieval from image databases

Content-based image retrieval (CBIRa technique for retrieving images on the basis of automatically-derived features such as colour, texture and shape.The issues discussed are color variance, image compression and image database browsing.



Image compression, which is usually used on most images, leads to both processing overheads but also decreases retrieval performance.

Sensor responses that describe an image taken with a device such as a digital color camera:

px = ex · nx ω Sx(λ)E(λ)R(λ)dλ  
  
Where
λ: Wavelength
p: A 3-vector of sensor responses (RGB pixel values)
Sx: The surface reflectance at location x
E: The spectral power distribution of the illumination
R: The 3-vector of sensitivity functions of the device.

Exploring systems give an outline of the image database to allow for intuitive navigation through the image collection. This is particularly in the option when images are arranged according to mutual similarity, that is, visually matching images are located close to each other in the browsing interface.

Hue Sphere browsing system provides a very efficient and effective approach to image database browsing. A spherical visualization space is used, since users are familiar with the concept of a globe and know how to navigate on one. Each image is identified by its median color. However,instead of using RGB color space, HSV space is used as people find it more intuitive.

H values and V values are then mapped to longitude and latitude values which point to the image thumbnail on the Hue sphere.

This approach is quick as the co-ordinates are extracted directly from the images and more efficient and faster than methods such as multi-dimensional scaling which rely on computationally expensive iterative convergence procedure.

References:

http://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=5466891

Comments