| | |
Summary: USING TEXTURE IN IMAGE SIMILARITY AND RETRIEVAL
SELIM AKSOY AND ROBERT M. HARALICK
Intelligent Systems Laboratory, Department of Electrical Engineering,
University of Washington, Seattle, WA 98195-2500, U.S.A.
E-mail: {aksoy,haralick}@isl.ee.washington.edu
Texture has been one of the most popular representations in image retrieval. Our
image database retrieval system uses two sets of textural features, first one being
the line-angle-ratio statistics which is a macro texture measure that uses a texture
histogram computed from the spatial relationships of intersecting lines as well as
the properties of their surroundings, second one being the variances of gray level
spatial dependencies computed from co-occurrence matrices as micro texture mea-
sures. This paper also discusses a line selection algorithm to eliminate insignificant
lines and statistical feature selection methods to select the best performing subset
of features to adjust the parameters of the feature extraction algorithms. Average
precision is used to evaluate the retrieval performance in comparative tests with
three other texture analysis algorithms. Experiments on a database of approxi-
mately 10,000 images show that low-level textural features can help in grouping
images into semantically meaningful categories and our method is fast and effective
with an average precision of 0.73 when 12 images are retrieved.
1 Introduction
|
