Texture Classification Using Dense Micro-Block Difference

Abstract

This paper is devoted to the problem of texture classification. Motivated by recent advancements in the field of compressive sensing and keypoints descriptors, a set of novel features called dense micro-block difference (DMD) is proposed. These features provide highly descriptive representation of image patches by densely capturing the granularities at multiple scales and orientations. Unlike most of the earlier work on local features, the DMD does not involve any quantization, thus retaining the complete information. We demonstrate that the DMD have dimensionality much lower than Scale Invariant Feature Transform (SIFT) and can be computed using integral image much faster than SIFT. The proposed features are encoded using the Fisher vector method to obtain an image descriptor, which considers high-order statistics. The proposed image representation is combined with the linear support vector machine classifier. Extensive experiments are conducted on five texture data sets (KTH-TIPS, UMD, KTH-TIPS-2a, Brodatz, and Curet) using standard protocols. The results demonstrate that our approach outperforms the state-of-the-art in texture classification.

Texture classification using block intensity and gradient difference (BIGD) descriptor

“…To investigate the structural features of image patches, we randomly select multiple pairs of smaller square regions with various scales. Features extracted from these region pairs encode the local structures of patches at different spatial granularities and orientations and have higher robustness to noise than those extracted from raw pixels [15]. For simplicity, we specify these smaller square regions within the image patch as "blocks".…”

“…λ = 1/(#classes × #training images) and maximum number of iterations (e.g. 100×#training images), we use the same setting as DMD [15] for a fair comparison. To guarantee fair comparisons between the proposed method and state-of-the-art ones, we keep the parameter setting unchanged for all databases unless we specify it.…”

“…In contrast, as Table 3 shows, for the KTH-TIPS-2a database, Table 4. For consistency, we set parameter pair (L, s) for the Brodatz database as (15,3) and for the KTH-TIPS-2a database as (13,2), which correspond to the best classification performance in Tables 2 Table 4, we notice that the classification accuracy of two datasets keeps growing with the increase of K. Although K with a value larger than 128 may correspond to higher classification accuracy, for computational efficiency, we set K as 128 for all experiments unless mentioned otherwise.…”

“…Distinctive and robust representation of texture is the key for various multimedia applications such as image representation [1], texture retrieval [2], face recognition [3], image quality assessment [4,5], image/texture segmentation [6], dynamic texture/scene recognition [7,3], texture/color style transfer [8], and seismic interpretation [9]. Texture descriptors [10,11,12,13,14,15,16], which are robust against rotations and translations of images, are able to provide discriminative features.…”

“…Mehta et al [15] proposed a novel descriptor called dense micro-block difference (DMD). Both methods composed of real-valued intensity differences instead of binary codes of different micro-blocks in local patches.…”