A New Texture Direction Feature Descriptor and Its Application in Content-Based Image Retrieval
Local Binary Pattern (LBP) has been widely used in texture analysis and content-based image retrieval (CBIR). LBP encodes the relationship between the referenced pixel and its surrounding neighbors by computing gray-level variation. However, LBP is unable
- PDF / 628,651 Bytes
- 9 Pages / 439.37 x 666.142 pts Page_size
- 49 Downloads / 218 Views
A New Texture Direction Feature Descriptor and Its Application in Content-Based Image Retrieval Yu Xia, Shouhong Wan and Lihua Yue
Abstract Local Binary Pattern (LBP) has been widely used in texture analysis and content-based image retrieval (CBIR). LBP encodes the relationship between the referenced pixel and its surrounding neighbors by computing gray-level variation. However, LBP is unable to reflect the spatial distribution information of gray variation direction in the whole image. Therefore, in this paper, we propose a new texture direction feature descriptor to extract the spatial distribution information of gray-level variation between pixels. After the calculation of the gray variation pattern on different directions, we construct the statistic histograms of pattern pairs between the referenced pixel and its neighbor pixels. The performance of the proposed feature descriptor is compared with different methods using two benchmark image databases. Performance analysis shows that the proposed feature descriptor improves the retrieval precision rate, as well as the recall rate both in texture and natural scene images. Keywords CBIR
Local binary pattern Texture analysis Feature direction
14.1 Introduction With the increasing demand of image retrieval on large image database, there exists a strong need for developing an efficient technology that can automatically search the desired image from the huge database. Content-based image retrieval Y. Xia S. Wan (&) L. Yue University of Science and Technology of China (West Campus), Hefei, China e-mail: [email protected] Y. Xia e-mail: [email protected] L. Yue e-mail: [email protected]
A. A. Farag et al. (eds.), Proceedings of the 3rd International Conference on Multimedia Technology (ICMT 2013), Lecture Notes in Electrical Engineering 278, DOI: 10.1007/978-3-642-41407-7_14, Springer-Verlag Berlin Heidelberg 2014
143
144
Y. Xia et al.
(CBIR) is one of the most useful solutions for such applications. CBIR refers to using the visual contents of images such as color, texture, shape, space layout, semantics to describe the image, and designing proper similarity measurement of image features to retrieval images which a user may be interested in. Local Binary Pattern (LBP) has emerged as a shining spot in the field of texture classification and retrieval. Ojala et al. [1] proposed LBP, which has been improved to rotational invariant version for texture classification (Rotational invariant LBP, RILBP) [5]. Various extensions of the LBP, such as uniform LBP (ULBP) [4], completed LBP (CLBP) [3], dominant LBP (DLBP) [2], local ternary pattern (LTP) [6] are proposed for texture classification. ULBP re-encoded the 256-dimensional LBP value and normalized the code result as the new feature vector. Using the same calculation method of LBP, CLBP encoded the gray value variation magnitude as well. CLBP obtained the texture feature which can reflect gray value variation information. Subrahmanyam et al. [7]. proposed the local maximum edge binary pattern (LME
Data Loading...
