Using down-sampling for multiscale analysis of texture images

摘要

This work proposes the study of a simple yet efficient strategy to accomplish multiscale analysis of gray level images. The method is applied to the classification of texture/material images. The proposal employs histograms of down-sampled versions of the initial image to compose a feature vector. A statistical model based on Markov random fields is also developed and employed to explain how the information conveyed by the proposed descriptors can express degrees of homo/heterogeneity in the image. This is widely known to be a fundamental feature in texture analysis. The accuracy of our descriptors is compared to several state-of-the-art approaches and the achieved results confirm our expectation that a straightforward method can be efficiently employed even in the recognition of objects in large and complex databases. (C) 2019 Elsevier B.V. All rights reserved.