Paper Info
Title
Multi-scale region perpendicular local binary pattern: an effective feature for interest region description
Abstract
This paper proposes the perpendicular local binary pattern (PLBP) for efficiently describing textures in an interest region. Its novelty is two-fold: (1) the candidate generation scheme provides a set of patterns for each pixel, instead of conventionally assigning one pattern per pixel, and (2) an adaptive threshold based on the image contrast of a region is used. These modifications successfully enhance the robustness of PLBP to Gaussian noise as well as in near-uniform regions. We introduce the novel multi-scale region PLBP descriptor, which adopts the PLBP as its core feature. It defines multiple support regions from an interest point, sequentially performs ring-shaped and intensity order-based segmentations on each region, and pools PLBPs to corresponding segments. These steps are controlled easily by a set of parameters, thus offering high flexibility. Experimental results on challenging benchmarks, including three datasets of image matching and two datasets of object recognition, demonstrate the effectiveness of the proposed descriptor in handling common photometric and geometric transformations. It significantly improves the robustness, compared with current state-of-the-art descriptors, while maintaining a reasonable operational cost.
Year
DOI
Venue
2015
10.1007/s00371-014-0934-5
The Visual Computer: International Journal of Computer Graphics
Keywords
Field
DocType
perpendicular,local binary pattern
Computer vision,Perpendicular,Pattern recognition,Computer science,Image matching,Transformation geometry,Local binary patterns,Robustness (computer science),Artificial intelligence,Pixel,Gaussian noise,Cognitive neuroscience of visual object recognition
Journal
Volume
Issue
ISSN
31
4
1432-2315
Citations 
PageRank 
References 
5
0.40
32
Authors
2
Name
Order
Citations
PageRank
Thao-Ngoc Nguyen150.40
Kazunori Miyata216141.73