ABSTRACT
Image retrieval refers to the process of finding out the image with specified characteristic or containing specified content in the set of target image according to the description of image content.[1] Spatial relationship expresses the relationship between each part within the image, which is a very obscure definition. More often than not, they cannot be clearly described with a definite expression; instead, they are often defined by a group of mutually related but contradictory conditions, with each condition being met to varying degrees under specific circumstances.[2] Generally speaking, the frequently-used spatial relationship covers orientation relationship (also referred to as projection relationship) and topological relationship. Both the former and the latter can also be described in natural language. A kind of characteristic which has certain particular connection with spatial relationship can be called structural characteristic.[3] The expression of spatial relationship can be divided into two types: the expression based on target and that based on relationship.[4] Starting from the non-linear characteristic of images, such scholars as Zhang Jianpei and Yangjing[5] study the location of correlation (information) of color space in images and then put forward a approach of image retrieval integrated color-space informations. Mei Chengli and others[6] proposed a mode to substitute the representation to image contents by image joint feature for single image feature and then introduced the indexing method of spatial data structure after defining the similarity of images in the Euclidean space. Considering the estimated high dimension characteristic of the image joint feature, presented
The average distance from pixel with color k to all the pixels with color k is defined as follows:
m k i C k
( , ) ( )
( , ) ( )
=
−
∑1 1 1
(2)
This average distance can be regarded as a characteristic of Pixel pi, which expresses the degree of distance from Pixel pi to other pixels with color k.
