On ratio-based color indexing

Abstract

The color ratio approach to indexing has been found to be robust and effective in indexing image and video databases, in different color spaces, and when using transformed color features, such as those from the Karhunen-Loeve transform (KLT) or the discrete cosine transform (DCT). However, the reason for the superior performance of the color ratio model, especially on different color spaces or with transformed color features has, at best, been speculative. This paper develops a generalized form for the color ratio model, based on which we characterize the general distribution of the color ratios. From the distribution, we present a theory that explains and supports the performance of the color ratio approach in image and video indexing. It is shown that the same theory accounts for its effectiveness in different color spaces and in the transform domain. Some general problems encountered in using the original retinex lightness algorithm, and some other issues specific to ratio-based color indexing are discussed in the light of the theory. Results are presented which show that the proposed theory is supported by empirical evidence.

A New Approach for CBIR Feedback based Image Classifier

“…There exist several common techniques for measuring the distance (dissimilarity) between two N-dimensional feature vector f and g. Each metric has some important characteristics related to an application. The following are the most important metrics used in the literature [10], [11].…”

“…A typical user-system session is as follows: A user presents an image query to the system where upon the system retrieves a fixed number of images using a default Similarity metric [11]. The user then rates each returned result with respect to how useful the result is for his or her retrieval task at hand.…”

Indexing iris images using the Burrows-Wheeler Transform

“…To overcome this limitation, we adapt the color ratio model [16] to generate illumination-invariant versions of the iris images.…”

An efficient retinex-like brightness normalization method for coding camera flashes and strong brightness variation in videos

“…These methods can generally be classified into two classes: (i) those which preserve pixel values and extract information for matching to facilitate motion estimation, (ii) those which modify pixel values in order to bring the brightness levels of two images, having different lighting conditions, close to each other. Algorithms of class (i) include motion estimation methods using mutual information [3], matching methods using the direction of the image gradient [4,5] and the extraction and matching of feature points [6][7][8]. Generally speaking, mutual information based approaches make use of a two-variable joint histogram of the matching images and measure the distribution of the histogram to determine if a match is found.…”

“…Similarly, some researchers have proposed extracting the multiscale oriented filters output of some feature points to form a feature vector for retrieval applications [6]. Funt and Finlayson [7] and Adjeroh and Lee [8] proposed extracting feature vectors from ratio images which are independent of illumination. However, most of the techniques focus on matching, recognition and retrieval applications.…”