Information Technology Reference
In-Depth Information
A Fast Detection Method for Bottle Caps
Surface Defect Based on Sparse Representation
Wenju Zhou 1 , 2 ,MinruiFei 1 ,HuiyuZhou 3 ,andZheLi 1
1 School of Mechatronical Engineering and Automation,
Shanghai University, Shanghai, 200072, China
mrfei@staff.shu.edu.cn
2 School of Information and Electronic Engineering,
Ludong University, Yantai, China
zhouwenju2004@126.com
3 The Institute of Electronics, Communications and Information Technology,
Queen's University Belfast, United Kingdom
h.zhou@ecit.qub.ac.uk
Abstract. A machine-vision-based system is developed for detecting de-
fects occurring on the surface of bottle caps. This system adopts a novel
algorithm which uses circular region projection histogram (CRPH) as the
matching feature. A fast algorithm is proposed based on sparse represen-
tation for speed-up searching. The non-zero elements of the sparse vector
indicate the defect size and position. Experimental results show that the
proposed method is superior to the orientation code method (OCM) and
has promising results for detecting defects on the caps' surface.
Keywords: detect defection, bottle cap, circular region projection his-
togram(CRPH), sparse representation.
1 Introduction
Cap is a very important part of the bottling product packaging. The pattern of
cap surface normally includes a company logo, but it is likely to be polluted such
as surface scratch, distortion, stains, printing deviation, and other ill-defined
faults during the production. Therefore, inspection of bad caps is of crucial
importance for quality control. Much work has been done on the subject of defect
detection, such as fabric [1], lumber [2], and bottling industries [3]. However, few
studies have addressed the inspection of the surface of bottle caps. The goal of
this paper focuses on the problem of defect detection of bottle caps' surface.
Most of the automated visual inspection systems for complicated
textured-surfaces generally attempt to identify defects by building adequate tem-
plates of features representationusing sample images. This representationis called
feature dictionary. Detection accuracy is dependent on how adequate and general
the dictionary is. Generally, the selection of an adequate feature set in the train-
ing process requires the help of complicated classifiers such as Bayes [4], maximum
likelihood[5], and neuralnetworksforclassifyingsamplefeatures andtemplate fea-
tures. The entire processis of high computationalcomplexity and time-consuming.
 
Search WWH ::




Custom Search