Band weighted colour histograms for image retrieval
A method of image retrieval from a target image collection, including segmenting a query image into two or more bands, obtaining weighted color histogram vectors for the two or more bands in the query image, and obtaining weighted color histogram vectors for two or more bands in a target image. A distance measurement is determined between the query image and the target image using the weighted color histogram vectors from corresponding bands in the query image and the target image. Bands in an image can be groups, strips, sections, regions, or the like, and can be linear bands, rectangular bands, circular bands, or the like. The bands are preferably, though not necessarily, concentric or otherwise aligned and any number of bands can be utilized. Also, content based image retrieval is provided that incorporates use of varying photo composition in different regions as a technique for improving accuracy of retrieval.
1. A computer-implemented method of image retrieval from a target image collection, the steps performed by a computer system and comprising:
segmenting a query image into two or more bands;
obtaining weighted colour histogram vectors for the two or more bands in the query image;
obtaining weighted colour histogram vectors for two or more bands in a target image; and,
determining a distance measurement between the query image and the target image using the weighted colour histogram vectors from corresponding bands in the query image and the target image;
wherein the distance measurement (D) between the query image (I q ) and the target image (I m ) is determined using:
D
(
I
q
,
I
m
)
=
∑
i
=
1
N
w
i
·
d
i
(
x
qi
,
x
ni
)
where:
d i is the individual distance measurement for selected corresponding bands between a colour histogram x q of the query image and a colour histogram x n of the target image; and
w i is a weight applied to the i th colour histogram x.
2. The computer-implemented method of claim 1 , wherein the two or more bands are two or more groups, strips, sections or regions.
3. The computer-implemented method of claim 1 , wherein at least one of the two or more bands is a rectangular band or a circular band.
4. The computer-implemented method of claim 1 , wherein the two or more bands are concentric.
5. The computer-implemented method of claim 1 , wherein there are three or more bands.
6. The computer-implemented method of claim 1 , wherein a central band of an image is allocated a larger weighting than an outer band of the image.
7. The computer-implemented method of claim 1 , wherein the two or more bands are of different widths.
8. The computer-implemented method of claim 1 , wherein the two or more bands are formed of at least two different geometries of bands.
9. The computer-implemented method of claim 1 , wherein the two or more bands are offset from the center of the image.
10. The computer-implemented method of claim 1 , wherein the query image includes a facial image.
11. The computer-implemented method of claim 1 , wherein retrieved images are ranked in order of display.