Query by example using invariant features from the double dyadic dual-tree complex wavelet transform.
ABSTRACT Widespread use of digital imagery has resulted in a need to manage large collections of images. Systems providing query by example (QBE) capability offer improved access to contents of image libraries by retrieving matches to a query image. Texture is an important feature to consider in the matching process. However, standard approaches often employ a texture feature that is scale and rotation specific, and may not perform well in libraries containing images with scaled or rotated matches to the target query. A novel approach for generating scale and rotation invariant texture features from an extension of the Dual-Tree Complex Wavelet Transform (DT-CWT) is presented herein for use in region-based QBE. An experimental comparison reveals an improved ability of the new technique in retrieving relevant images over the standard approach.
- SourceAvailable from: rpi.edu[show abstract] [hide abstract]
ABSTRACT: In this paper we present the results of two experiments. The first is on the categorization of texture words in the English language. The goal was to determine whether there is a common basis for subjects' groupings of words related to visual texture, and if so, to identify the underlying dimensions used to categorize those words.Eleven major clusters were identified through hierarchical cluster analysis, ranging from ‘random’ to ‘repetitive’. These clusters remained intact in a multidimensional scaling solution. The stress for a three-dimensional solution obtained through multidimensional scaling was 0.18, meaning that 82% of the variance in the data is explained through the use of three dimensions. It appears that the major dimensions of texture descriptors are repetitive versus nonrepetitive; linearly oriented versus circularly oriented; and simple versus complex.In the second experiment we measured the strength of association between texture words and texture images. The goal was to determine whether there is any systematic correspondence between the domains of texture words and texture images. Pearson's coefficient of contingency, a measure of the strength of association, was found to be 0.63 for words corresponding to given images and 0.56 for images corresponding to given words. Thus the texture categories in the verbal space and those in the visual space are strongly tied.In sum, our two experiments show (a) that despite the tremendous variety in the words we have to describe textures, there is an underlying structure to the lexical space which can be derived from the experimental data; and (b) that the association between a category of words and a category of images was strongest when both categories represent the same underlying property. This suggests that subjects' organizations of texture terms are systematically tied to their organization of texture images.Cognitive Science. 01/1997;
- [show abstract] [hide abstract]
ABSTRACT: In this paper we propose a new texture segmentation technique that produces segmentation results which more closely match the manual segmentation that would be performed by a human operator. To perform this type of segmentation, we propose a new texture feature based on the double dyadic dual-tree complex wavelet transform (D<sup>3</sup>T-CWT) which provides the ability to analyse a signal at and between dyadic scales. This new texture feature is invariant to shift, rotation and scale and hence can group the texture features in a single object (which may have different sizes and orientations) into a single more meaningful segment. When compared with other texture segmentation approaches, the proposed approach provides segmentation results which more closely match the semantically meaningful objects in the scene.Acoustics, Speech and Signal Processing, 2007. ICASSP 2007. IEEE International Conference on; 05/2007 · 4.63 Impact Factor
- [show abstract] [hide abstract]
ABSTRACT: Several approaches to multichannel filtering for texture classification and segmentation with Gabor filters have been proposed. The rationale presented for the use of the Gabor filters is their relation to models for the early vision of mammals as well as their joint optimum resolution in time and frequency. In this work we present a critical evaluation of the Gabor filters as opposed to filter banks used in image coding--in both full rate and critically sampled realizations. In the critically sampled case, tremendous computational savings can be realized. We further evaluate the commonly used octave band decomposition versus alternative decompositions. We conclude that, for a texture segmentation task, several filters provide approximately the same results as the Gabor filter and, most important, it is possible to use subsampled filters with only a modest degradation in segmentation accuracy--realizing considerable computational savings.Optical Engineering 01/1994; 33(8):2617-2625. · 0.88 Impact Factor