A Fast Detection Method for Bottle Caps Surface Defect Based on Sparse Representation - Intelligent Computing for Sustainable Energy and Environment - page 77

In the meantime, the production of caps is very fast. Thereby, a high-speed detec-
tion method is urgently needed. In this paper, we propose a novel circular region
projection histogram (CRPH) method and a fast detection algorithm based on
sparse representation for detecting defects of bottle caps' surface.
The circular region projection histogram method is inspired by the orientation
code histogram method published in [6], and the fast detection algorithm is sim-
ilar as sparse decomposition mothed in [7]. In literature [7], text and piecewise
smooth contents in an image are separated into two different images. Different
dictionaries were used for different contents, such as a dictionary of biorthogo-
nal wavelet transforms (OWT) for piecewise smooth contents and a dictionary
of discrete cosine transform (DCT) for texture contents. In our method, the
image center is firstly located, and then the appropriate radius circle range is
extracted as the template region of interesting (ROI). The ROI is projected as
histograms along different directions. The histograms are the arrays that model
the true distribution by counting the occurrences of pixel values that fall into
each bin. These arrays are regarded as atoms that compose the template dic-
tionary. Secondly, the sample cap surface image is captured whilst the sample
ROI is extracted. The sample histograms at vertical and horizontal directions
are computed by projecting the ROI. Lastly, the defect can be found through
matching the atoms in the template dictionary to the sample histogram using a
developed sparse representation method. The experimental result shows that the
CRPH method and the developed sparse representation algorithm are effective
for solving the rotation match.
The rest of this paper is organized as follows. Section 2 reviews the challenges
and corresponding solutions based on sparse representation. Section 3 discusses
the processing of extracting ROI and introduces a novel method CRPH for
feature for matching, while the template dictionary and the defect dictionary are
built. The fast match algorithm is proposed in section 4. Experimental results
and their analysis are shown in section 5 and section 6 concludes the paper with
a summary of the proposed work and discussions.

2 Sparse Representation Method

Suppose that an arbitrary bottle cap has a surface image X which contains a
template pattern marks X t and a defect component X d . As such, the bottle cap
surface can be denoted as follows,

X = X t + X d . (1)

The defect detection is a process of separating the template image X t and defect
image X d from the sample image X . We can use a sparse representation to solve
this equation.
The template image X t only contains original patterns which is flawless.
Sparse decomposition matrix D t ∈

M N×L
t

is written as

X t = D t α t , (2)

where α t isthetemplatesparsefactor. D t is the template dictionary.