OkonomiDev
Friday, March 21, 2003
      ( 8:44 AM ) Matt  

Image Database


There are two important features that the image database needs to have are sorting by similarity of wavelet signature, and searching for similar wavelet moments. These two functions are very similar, so it shouldn't be hard to come up with a database structure which fascilitates both.


Before we go any further, let me give some definitions.


  • wavelet moment - the collection of wavelet coefficients describing one location or point.

  • moment magnitude - the average magnitude of the coefficients in a moment.

  • significant - A moment is more significant if it has a greater magnitude.

  • wavelet signature - a small collection of the most significant moments in an image.


In order to facilitate searching through all wavelet moments, we'd need to store all the wavelet moments. The number of coeficients in a wavelet moment is 3 * lg m * lg n, where m and n are the width and height of the image. So, the number of coefficients per image is 3*m lg m*n lg n. Storing this data raw for a 1MP image would take up 100 times as much space as storing the image raw. Now, everyone says that hard drives are big these days, but they certainly aren't big enough for that.


We need to come up with a more efficient way to store these things.

# -




Comments: Post a Comment
archives:


Dreams I have...

Powered by Blogger
Feel free to e-mail me.

free hit counter