hello,
can you help me with algorithm and code logic of color autocorrelogram of an image...
thanks a lot..

I want to implementation cbir color autocorrelogram code in matlab and algorithm.
Could you advise me how to build the algorithm with the reasonable logic?
Thanks a lot.

cbir color autocorrelogram code in matlab and algorithm

ABSTRACT

Much of the information stored in digital libraries will contain either images or video, which is difficult to search or browse. Automatic methods for searching image collections make wide use of color histograms, because they are robust to large changes in viewpoint, and can be computed trivially. However, color histograms fail to incorporate spatial information, and therefore tend to give poor results. We have developed several methods for combining color information with spatial layout, while retaining the advantages of histograms. One technique computes the distribution of a given color as a function of the distance between two pixels. The resulting method, which we call a color correlogram, has proven to be quite effective even with very coarsely quantized color information. Another method computes joint histograms of local properties, thus dividing pixels into classes based on both color and spatial properties. Experiments with a database of over 200,000 images demonstrate that these measures perform significantly better than color histograms, especially when the number of images is large.
INTRODUCTION

One of the primary challenges in digital libraries is the problem of providing intelligent search mechanisms for multimedia collections. While there are good tools for searching text collections, images are much more difficult. If the images are annotated by hand, a textual search can be used; however, this approach is too labor-intensive to scale up with large digital libraries. Automated methods for searching large image collections are therefore necessary. This in turn requires simple and effective image features for comparing images based on their overall appearance. Color histograms are widely used, for example by [QBIC], [Chabot] and [Photobook]. The histogram is easy to compute and is insensitive to small changes in viewing positions. A histogram is a coarse characterization of an image, however, and images with very different appearances can have similar histograms. For example, the images shown in figure 1 have similar color histograms. When image databases are large, this problem is especially acute.Since histograms do not include any spatial information, recently several approaches have attempted to incorporate spatial information with color [Hsu, Stricker, Smith]. These methods, however, lose many of the advantages of color histograms. In this paper we describe methods for combining color information with spatial layout while retaining the advantages of histograms. One method computes the spatial correlation of pairs of colors as a function of the distance between pixels. We call this feature a color correlogram (the term ``correlogram'' is adapted from spatial data analysis [Upton]) Another approach is based on computing joint histograms of several local properties. Joint histograms can be compared as vectors, just as color histograms can. However, in a color histogram any two pixels of the same color are effectively identical. With joint histograms, pixels must share several properties beyond color. We call this approach histogram refinement. The methods we describe are easy to compute, and they produce concise summaries of the image.

We will next describe color correlograms and histogram refinement (for details see [Huang] and [Pass]. We have evaluated these methods using a large database of images, on tasks with a simple, intuitive notion of ground truth. The experimental results that we present show that our methods are significantly more efficient than color histograms.