VQ Based Image Retrieval Using Color and Position Features

Abstract

We present a new lower complexity approach for content based image retrieval based on a relative compressibility similarity measure using VQ codebooks employing feature vectors based on color and position. In previous work we have developed a system that employs feature vectors that are a combination of color and position. In this paper, we present a new approach that decouples color and position. We present this approach as two methods. The first trains separate codebooks for color and position features, eliminating the need for potentially application specific feature weightings during training. The second method achieves nearly the same performance at greatly reduced complexity by partitioning images into regions and training high-rate TSVQ codebooks for each region (i.e., position information is made implicit). Features extracted from query regions are encoded with the corresponding database region codebooks. The maximum number of codewords that a database region codebook may contain is determined at runtime and is a function of the query features. Region codebooks are then pruned appropriately before encoding query features. Experiments performed on the COREL image database show this new approach to provide almost equivalent retrieval precision to our previous method of jointly trained codebooks (and an improvement over previous methods) at much lower complexity. BackgroundWith the recent proliferation of digital images, there is a need for information systems that can organize and store images using models which support content based queries. In the query-by-example setting (Eakins and Graham [4]), a user presents the system with a query image and the system responds by retrieving a set of database images with (visually) similar content. Given the discriminative power of color features and the simple histogram model, global color histograms which are relatively invariant to spatial transformations such as translation and rotation, have been effectively used

A Comparative Analysis of Feature Sets for Image Classification Using Back Propagation Neural Network

“…There is no definite agreement on the selection of features for best results. The classifiers used vary from the rule and fuzzy rules based [9], support vector machines [13], [26]; ART [13], SOM [22], and vector quantization [27], [28]. The most popular and successful classifier described by many researchers has been the back propagation neural network [5], [6], [10], [12]- [14] which has been selected for evaluation of the different types of features.…”

A fast compression-based similarity measure with applications to content-based image retrieval

“…Among these, the Minimum Distortion Image Retrieval (MDIR) by Jeong and Gray outperforms previous techniques based on histogram matching, by fitting to the training data Gaussian Mixture Models, later used to encode the query features and to compute the overall distortion [24] [32]. Daptardar and Storer introduce then a similar approach using VQ codebooks and mean squared error (MSE) distortion, decoupling to some degree spectral and spatial information by training separate codebooks in different regions of the images, outperforming in turn MDIR: we refer to their methodology as Jointly Trained Codebooks (JTC) [33].…”

Compression-Based Tools for Navigation with an Image Database

“…In past work, we have developed a color-based retrieval system based on a form of differential compression where similarity is measured by employing vector quantization (VQ) to compress one image in terms of a codebook based on a second image; that is, similar images tend to compress well with respect to the other as compared to less similar ones (Daptardar and Storer [1]). This kind of approach has been successful in the past for testing similarity of text (e.g., Shapira and Storer [2,3]), although it is a much different problem for imagery.…”