Making texture descriptors invariant to blur.

Andreas Uhl,Michael Gadermayr

doi:10.1186/s13640-016-0116-7

Abstract

Besides a high distinctiveness, robustness (or invariance) to image degradations is very desirable for texture feature extraction methods in real-world applications. In this paper, focus is on making arbitrary texture descriptors invariant to blur which is often prevalent in real image data. From previous work, we know that most state-of-the-art texture feature extraction methods are unable to cope even with minor blur degradations if the classifier’s training stage is based on idealistic data. However, if the training set suffers similarly from the degradations, the obtained accuracies are significantly higher. Exploiting that knowledge, in this approach the level of blur of each image is increased to a certain threshold, based on the estimation of a blur measure. Experiments with synthetically degraded data show that the method is able to generate a high degree of blur invariance without loosing too much distinctiveness. Finally, we show that our method is not limited to ideal Gaussian blur.

Highlights

For some decades, texture classification [1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16] has been a fundamental challenge in image processing
Blur measures are usually not built to measure blur in textured images but rather in natural scenes. These metrics are constructed to measure the perceptual image degeneration and not the Gaussian σ. As it is not clear, which properties of a blur measure are important in case of our scenario, in a first step, we investigate them with respect to two prediction rates
By equalizing the blur level, a high degree of invariance can be achieved without losing too much distinctiveness, which is of high relevance for practical usage

Summary

Introduction

Texture classification [1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16] has been a fundamental challenge in image processing. Texture descriptors have to capture all intrinsic image properties. These are properties that contain distinctive information (for discrimination) and do not depend on the image acquisition conditions. Extrinsic properties (i.e., properties that vary with different acquisition conditions) should not be captured, in order to maintain invariance to specific properties. We focus on blur which is usually caused by defocus, motion, or chromatic aberrations. In case of good image acquisition conditions blur can mostly be prevented effectively, in many realworld scenarios, this degradation still features a present problem. It is quite difficult to adjust the distance between the lens and the surface, which is a source for defocus aberrations. The permanent activity of the bowel in combination with a difficult handling of the endoscope is a source for motion

Methods

Results

Conclusion