Using no‐parameter statistic features for texture image retrieval

Abstract

Purpose

The purpose of this paper is to analyze the spectrum influence between radon transform and log‐polar transform when rotation and scale effect is eliminated. The average retrieval performance of wavelet and NSCT with different retrieval parameters is also studied.

Design/methodology/approach

The authors designed a multi‐scale and multi‐orientation texture transform spectrum, as well as rotation‐invariant feature vector and its measurement criteria. Then a new two‐level coarse‐to‐fine rotation and scale‐invariant texture retrieval algorithm based on no‐parameter statistic features was proposed. Experiments on VisTex texture database show that the algorithm proposed in this paper is appropriate for main orientation capturing and detail information description.

Findings

According to the experiments results, it was found that the combination of this two‐level progressive retrieval strategy and multi‐scale analysis method can effectively improve retrieval efficiency compared with traditional algorithms and ensure a high precision as well.

Originality/value

The paper presents a novel algorithm for rotation and scale‐invariant texture retrieval.

Citation

Zhu, X. and Shao, Z. (2011), "Using no‐parameter statistic features for texture image retrieval", Sensor Review, Vol. 31 No. 2, pp. 144-153. https://doi.org/10.1108/02602281111110004