Learning How to Extract Rotation-Invariant and Scale-Invariant Features from Texture Images

Abstract

Learning how to extract texture features from noncontrolled environments characterized by distorted images is a still-open task. By using a new rotation-invariant and scale-invariant image descriptor based on steerable pyramid decomposition, and a novel multiclass recognition method based on optimum-path forest, a new texture recognition system is proposed. By combining the discriminating power of our image descriptor and classifier, our system uses small-size feature vectors to characterize texture images without compromising overall classification rates. State-of-the-art recognition results are further presented on the Brodatz data set. High classification rates demonstrate the superiority of the proposed system.

Supervised pattern classification based on optimum‐path forest

“…Color images are also represented by color histograms (CHIST) [57] and compared with L1 metric in the descriptor D 7 . Descriptor D 8 uses steerable pyramid decomposition to create texture features (TEX), which are compared by a rotationinvariant texture matching (RIM) [35]. Descriptor D 9 represents all feature vectors (OWN) already available in the datasets from B 4 to B 7 and D 10 represents the 2D-point (XY) features of the datasets from B 8 to B 11 (Table 1).…”

“…A pair (v, d) then describes how the samples of a dataset are distributed in the feature space. Therefore, we call (v, d) a descriptor and the experiments in Section 4 use shape [41], texture [35] and color [42] descriptors based on this definition.…”

“…The proposed method can be easily extended to semi-supervised classification, given that the optimum-path forest can include unlabeled non-prototype samples. Previous versions of it have also been published [33][34][35][36][37]. We have simplified the learning procedure with better results, corrected some mistakes, improved explanations and added several experiments using more datasets, baseline classifiers, and image descriptors based on texture, shape and color.…”

On the Training Patterns Pruning for Optimum-Path Forest

“…A pair (v, d) then describes how the samples of a dataset are distributed in the feature space. Therefore, we call (v, d) a descriptor and the experiments in Section 4 use shape [10], texture [6] and color [11] descriptors based on this definition.…”

“…The OPF classifier has some advantages with respect to Artificial Neural Networks using Multilayer Perceptron (ANN-MLP) [2] and Support Vector Machines (SVM) [3]: (i) one of them is free of parameters, (ii) they do not assume any shape/separability of the feature space and (iii) run training phase faster. Notice that the OPF classifier has been extensively used for several purposes, including laryngeal pathology [4] and oropharyngeal dysphagia detection [5], fingerprint classification [6] and satellite-based rainfall estimation [7]. Some specialist systems, such that image-based medical diagnosis systems, need to be constantly re-trained, allowing a better generalization performance.…”