Options
Texture-Based Food Recognition Using Homogeneous Kernel Mapping
Date Issued
2015
Date
2015
Author(s)
Chen, Wei-Zong
Abstract
Balanced nutrition diet has been gained more attention recently. To keep balanced diet, users have to record each meal for a long period. However, typing on cellphone is inconvenient. To solve this problem, we proposed a system which records what we eat by taking photo. In the system we proposed a simple but effective recognition method. First, for each channel, we extract local binary pattern, calculate descriptor histogram for each image patch. Then we concatenate channel histogram together. Second, we learn a visual dictionary, encode the patch using sparse coding, and aggregate descriptors with mean pooling. Finally we transform our feature vector obtained above with homogeneous kernel mapping, which approximate kernel support vector machine with linear model. Besides, we introduce a method to augment the dataset by flipping local binary pattern. The experiment shows that we have better accuracy than the method proposed by Chen et al. (2012), and reduce consumed time effectively.
Subjects
food recognition
image recognition
local binary pattern
sparse coding
homogeneous kernel mapping
Type
thesis
File(s)
No Thumbnail Available
Name
ntu-104-R02922088-1.pdf
Size
23.32 KB
Format
Adobe PDF
Checksum
(MD5):d005e14a6408330dc8d8b3f638e6e800