Defects Detection in Printed Circuit Boards by Separation and

Comparison of Chains

1ROMAN MELNYK, 2ROMAN KVIT

1Software Department,

Lviv Polytechnic National University,

12 Stepan Bandera St., Lviv, 79013,

UKRAINE

2Department of Mathematics,

Lviv Polytechnic National University,

12 Stepan Bandera St., Lviv, 79013,

UKRAINE

Abstract: - The known K-means clustering, flood-filling, and thinning algorithms are used to find coordinates

of contacts in PCB images. Images of different types and colors are considered. The clustering algorithm is

used to reduce the number of colors, to get uniform colors in the PCB image. The thinning algorithm is used to

build skeletons and find pixels of contacts. The flood-fill algorithm is used to mark and separate chains, defect

connectivity, and metal defects. Different subtraction operations are applied to original, transformed, and

distributed cumulative histogram images.

Key-Words: - PCB, defects, chain, contact, flood-filling, thinning, clustering, subtraction, histogram,

approximation

Received: April 24, 2022. Revised: April 18, 2023. Accepted: May 19, 2023. Published: June 26, 2023.

1 Introduction

There are many approaches for printed circuit board

defect detection in the published articles. They are

grouped into three classes. The first one unites

works describing artificial networks for fault

diagnostics. For example, the works, [1], [2], [3],

use neural network tools. Works of the second class

are based on subtraction algorithms as it is presented

in the papers, [4], [5], [6], where defects are

separated and classified. The article, [7], considers

defects detection based on a connected table of a

reference image. This table is often unknown,

and the coordinates of contacts are to be

determined by different methods.

Some works considering defects in PCB images

and the method of their extraction are given in the

surveys, [8], [9], [10]. In [11], different defect

methods are divided into groups of image

subtraction and feature extraction for the PCB

image control. The algorithms in [12], [13], use

parts of the board to increase the visibility of

defects. Simple techniques to divide a plane are

used. In the works, [14], [15], authors propose an

algorithm for machine vision inspection systems to

detect the defects short circuits, open circuits, and

depressions defects in a PCB image. The work, [13],

considers parallelization issues for sequential

algorithms of thinning. Algorithms of thinning,

clustering, and comparison are used in the work,

[16], to find defective contacts and traces.

In this article three approaches are considered:

different types of subtraction and reprocessing,

thinning and flood filling to select and separate

chains, measurement of traces to detect positions

and intensity of PCB defects. Algorithms of K-

means clustering and calculation of distributed

cumulative histograms are applied. PCB samples for

testing of algorithms are taken from [17], [18].

Many subtraction approaches give in result an

image that is controlled by the operator. He can miss

defective cases. The proposed approach is fully

automatic based on the consequence of the

developed algorithms.

Also, the subtraction operations are very

sensitive to the deviation of the sizes of contacts and

traces from the reference ones and give extra or

lacking pixels that do not influence the normal work

of the circuit.

WSEAS TRANSACTIONS on CIRCUITS and SYSTEMS

DOI: 10.37394/23201.2023.22.9

Roman Melnyk, Roman Kvit

E-ISSN: 2224-266X

63

Volume 22, 2023

2 Subtraction and Comparison

Formulas

2.1 Logical Subtraction Operations

A simple way to isolate defects in a manufactured

PCB image (denoted as

m

B

) is to subtract its binary

form

C

from the corresponding binary form

2r

B

of

the reference PCB image (denoted as

r

B

). But in

this case, along with defects, the result contains

manufacturing inaccuracies that often do not affect

the functionality of the scheme.

The

XOR

operations with pixels are as follows:

0^1 1,1^ 0 1, 0^0 0,1^1 0,   

where

0

denotes white and

1

denotes black or vice

versa.

Applying

XOR

operations to the binary images

of printed circuit boards considered below, the

resulting image with marked defects is obtained,

which is shown in Figure 1a. Manufacturing

inaccuracies are mixed with various defects. Some

defects are isolated from traces and contacts. It is

difficult to remove defects from such an image. It is

impossible to classify them on the received image.

The second approach involves attaching

inaccuracies to objects on the board. For this, a

modified

OR

logical operator is used. The

modification is that the result of the two operations

is marked with a red

0^1 1

(red),

1^0 1

(red),

0^0 0

,

1^1 1

,

where

0

denotes white and

1

denotes black (except

for those mentioned) or vice versa.

Then, in the image in Figure 1b, [4], objects are

black, and defects with manufacturing inaccuracies

remain red.

a b

Fig. 1: Difference between two PCB images: defects

and inaccuracies (a), objects, defects, and

inaccuracies (b)

2.2 Mathematical Subtraction Comparison

Formulas

To obtain a binary form segmenting and

thresholding procedures are required. It is a more

simple way to compare the input original images.

Two images are compared according to the

following comparison formulas

( , ) ( , )

| ( , ) ( , )| ,

( , ) (| ( , ) ( , )|,00)

| ( , ) ( , )| ,

rd

ed

r e d

ed

I x y I x y

I x y I x y Tol

I x y RGB I x y I x y

I x y I x y Tol



  



  

(1)

where

),( yxIr

is the pixel intensity of the resulting

image,

),( yxIe

is the pixel intensity of the first

image,

),( yxId

is the pixel intensity of the second

image,

Tol

and is the tolerance value for

controlling the difference between the pixel

intensity of the reference and controlled samples.

The resulting image obtained by formula (1) is

shown in Figure 2a. Pixels corresponding to positive

or negative intensity differences are marked in red.

To account for the sign of the pixel intensity

deviation, a refinement is used to specify the nature

of the deviation

( , ) (| ( , ) ( , )|,00)

( , ) ( , ) ,

( , ) (0,| ( , ) ( , )|,0)

( , ) ( , ) .

r e d

ed

r e d

ed

I x y RGB I x y I x y

I x y I x y Tol

I x y RGB I x y I x y

I x y I x y Tol



  



  

(2)

The resulting image obtained by formula (2) is

shown in Figure 2b. Pixels corresponding to positive

intensity differences are marked in red, and negative

intensity differences are marked in blue.

a b

Fig. 2: Difference between the two PCB images: red

defects and inaccuracies (a), red and blue defects

and inaccuracies (b)

All four approaches make it possible to isolate

defects along with manufacturing inaccuracies. The

latter distinguishes two groups of defects: excess

WSEAS TRANSACTIONS on CIRCUITS and SYSTEMS

DOI: 10.37394/23201.2023.22.9

Roman Melnyk, Roman Kvit

E-ISSN: 2224-266X

64

Volume 22, 2023

and missing bonding metal. The resulting image is

planned to be controlled by the operator. It is

difficult to interpret the results of the subtraction,

which type of defects has occurred because it

requires constant comparison with the input images.

In this way, the operator can miss defective cases.

Thus, a fully automated approach is preferable.

In addition, the subtraction operation is very

sensitive to the deviation of pin and trace sizes from

the reference image and sometimes gives extra or

missing metal pixels that do not affect the normal

operation of the circuit.

The mathematical comparison formulas are

auxiliary and designed to visualize defects in

individual circuits. It is difficult to interpret the

results of the subtraction, which type of defects has

occurred because it requires constant comparison

with the input images.

In this sense, the following approach is more

universal: for flood-filling, thinning, and other

algorithms, the sizes of tracks and contacts are

irrelevant. The main condition is the complete

display of the physical board with an image.

3 Pre-processing Algorithms

3.1 Distributed Cumulative Histogram

To build the image of a distributed cumulative

histogram two sets of ordinary histograms of the

number

N

of columns and

M

rows are calculated

( ) { ( )}, 0,255, 1, ;

( ) { ( )}, 0,255, 1, .

i i j

j ji

V c V c j i N

V r V r i j M

  

(3)

Then two distributed cumulative histograms

assets of frequency sums

0

( ) ( ), 0,255 , 1, ;

( ) ( ), 0,255 , 1, ,

i

j li

l

i

j li

l

V cc V r i j N

V cr V r i j M





  







  







(4)

where,

)(crVi

,

)(ccVj

are cumulative histograms in

rows and columns

)(cV j

i

,

)(rV i

l

are intensity

frequencies in a column and row,

,NM

are

columns and rows numbers.

The last equations are the mathematical models

of distributed cumulative histograms of an image.

An example of such processing (3)-(4) is given in

Figure 3, where the original image of “circles” is

shown in Figure 3a, and its DCH image is shown in

Figure 3b. The last image is covered by a grid

designed to measure intensity.

a

b

Fig. 3: The “circles” image (a), a 2-D image of

distributed cumulative histogram (b) covered by a

grid

The advantages of the DCH image are the

following: 1) the intensity of circles is automatically

compared, 2) the intensity of every circle is

automatically measured by horizontal lines of the

green grid, and the intensity interval

0 255

corresponds to the height of the grid, from down to

the top on the OY axis, 3) the DCH image reflects a

noise belonging to the surface of all circles and a

noise of a white background.

3.2 K-means Clustering Algorithm

The K-means clustering algorithm approximates the

PCB image by K segments of pixels with the same

intensity.

This algorithm assigns pixels into different

clusters to minimize the sum of distances between

the centroids and pixels within a cluster. The

criterion function is as follows:

2

1 0 1 0

| |, [ ] ,

m K m K

ik k ik k

i k i k

S w I I S w I I

   

   

   

(5)

where

1

ik

w

if a pixel of intensity

ir

I

belongs to

the cluster

k

, otherwise,

0

ik

w

,

k

I

is the

intensity of the i-th centroid.

The DCH image serves as control information

for processing and correction of the original image.

An example of such correction by the clustering

algorithm (5) is given in Figure 4. The original

image is clustered by seven clusters (circles and

WSEAS TRANSACTIONS on CIRCUITS and SYSTEMS

DOI: 10.37394/23201.2023.22.9

Roman Melnyk, Roman Kvit

E-ISSN: 2224-266X

65

Volume 22, 2023

background). The resulting image of the clustered

circles is given in Figure 4a. Its DCH image is

shown in Figure 4b.

a

b

c

Fig. 4: The clustered “circles” image (a), its DCH

image (b), and graphs of intensity (c)

4 Processing of Reference PCB Image

4.1 Clustering into a Binary Form

The first stage of the method includes the steps of

selecting circuit chains and marking their

coordinates. They include three algorithms:

transforming to the binary form by the clustering

algorithm, selection of chains by the flood-filling

algorithm, and searching of specific points with the

application of the thinning algorithm.

The application of the clustering algorithm is

explained below. For illustration, the image

histogram is calculated

( {( , )| ( , ) }.h i) card u v I u v i

Also, the cumulative histogram is calculated:





i

j

jhi)H

1

)((

,

where

),( vuI

is the pixel intensity, and

()hi

is

intensity frequency,

()Hi

is the accumulating

frequency for the given intensity.

The histogram of the input grey image from [19],

in Figure 5a has three maximum values (Figure 5b).

The first corresponds to the grey background and is

relatively large. This representation of the PCB

image is not suitable for the subtraction algorithm.

Therefore, the image must be transformed. The

approach from [19], applies three-interval

segmentation to select wires, pins, and holes. After

thresholding and redrawing, the selected segments

are used for a subtraction algorithm to find defects.

The weaknesses of this approach are the

necessary algorithms for determining thresholds, the

lack of defect types and coordinates, as well as the

lack of any quality characteristics of defects (area,

intensity, coordinates, etc.).

a b

Fig. 5: A PCB image (a) and its histogram (b)

A clustering algorithm is applied to the image to

obtain three segments (

3K

). The advantages of

the approach are as follows: 1) no pixel is deleted,

2) pixels are grouped naturally near their centroids,

3) segment colors are uniform and their values do

not affect the subsequent fill. The cluster image is

shown in Figure 6a. Visually changes are not

noticeable. Its usual histogram is shown in Figure

6b. For comparison, the very small graph is a red

histogram of the original input image of the PCB.

a b

Fig. 6: A clustered PCB image (a) and its histogram

(b) with original histogram

The DCH image distinctly illustrates how pixels

are distributed in the input and clustered images.

Figure 7b shows three groups of pixels

corresponding to contacts, traces, and backgrounds.

WSEAS TRANSACTIONS on CIRCUITS and SYSTEMS

DOI: 10.37394/23201.2023.22.9

Roman Melnyk, Roman Kvit

E-ISSN: 2224-266X

66

Volume 22, 2023

a b

Fig.. 7: DCH images for input (a) and clustered

images (b)

There are two ways of converting clustered

segments into a binary form: 1) image segmentation

by intensity levels from Figure 6b or Figure 7b,

separating the required parts, or 2) sequentially

filling the required part with white and all other

parts with black. The transformation of the clustered

image from Figure 4a is shown in Figure 8.

a b

Fig. 8: Two steps of flood-filling: a background

with white (a) and all other elements with black (b)

The simplest case is when the clustering

algorithm is applied to form two segments (

2K

).

Two images are obtained depending on redrawing

color. They are shown in Figure 9: two clusters in

Figure 9a and this image after redrawing in Figure

9b.

a b

Fig. 9: Two clusters in the PCB image: grey (a), and

after redrawing (b)

The binary PCB images in Figure 8b and Figure

9b are ready for further processing.

4.2 Separation of Connected Components

The reference image of the printed circuit board

consists of the background and various components

distributed on its surface. Components are isolated

contacts, traces, and chains (contacts connected to

traces). The actual PCB image may contain some

isolated metal extra surfaces. The purpose of the

developed algorithms is to separate these

components and compare them instead of

comparing whole images.

Hilditch's thinning algorithm, [8], works with

binary (black and white) images. This tool is very

important because it finds the pixel coordinates that

are accepted as chain contacts. This is useful for

locating, highlighting, and separating electrical

circuits as connected components on a board. These

are used as starting points for the filling algorithm to

build and select chains in the reference and real

PCB images for comparison.

Figure 10 shows that the skeleton image has

three types of elements: chains with specific points

placed (contacts and traces), circles without any

specific points, and isolated specific points of noise

or extra metal on the background. In general, the

last elements should not be placed on the

background because the image is made for the

template PCB. If noticed, they must be deleted from

the image in a manual or programmed way.

Fig. 10: Skeleton with endings and switches

Figure 10 shows that some circles in the skeleton

have no specific points. They correspond to contacts

without traces. It means that a simple way of

thinning does not allow us to separate these

components for comparison.

Therefore, another method is used. The circles

separated from the skeleton are shown in Figure

10a.

To illustrate the basic idea, the custom histogram

is calculated, but instead of the traditional

probability/intensity axes, it uses

probability/coordinates axes (

OX

or

OY

)

WSEAS TRANSACTIONS on CIRCUITS and SYSTEMS

DOI: 10.37394/23201.2023.22.9

Roman Melnyk, Roman Kvit

E-ISSN: 2224-266X

67

Volume 22, 2023

( ) {( , )| ( , ) ( ) ( )}, 0,1, ... ,h i card i v I i v I b I e i   

width (height), (7)

where

),( viI

is the pixel intensity,

()hi

is the

number of pixels, and

()Ib

,

()Ie

are borders of the

intensity interval of the counted pixels.

Histograms of black pixels in columns and rows

of circles are calculated. Figure 11b shows the

histogram for all circles by the axis

OX

. Figure 12

shows the increased image of one circle and its

histogram by the axis

OY

.

a

b

Fig. 11: Circles image (a), a histogram by the

OX

axis (b)

a b

Fig. 12: An image of one increased circle (a), and its

histogram by the

OY

axis (b)

Two histograms have extrema values. For one

circle every histogram has two maximums and one

minimum, which are found in the graph with W

points

max(min) max(min) ( ), 1h h i i W  

.

Three coordinates in the axis

OY

and three

coordinates in the axis

OX

give coordinates of four

points on the circle-skeleton. The process of their

finding is shown in Figure 13: a - finding of lines, b

- crossing lines. Blue lines correspond to minimum

values in two histograms and red lines correspond to

maximum values in two histograms. The

intersection of two different lines gives the specific

point on the circle-skeleton. They are four small

circles in Figure 13b.

a b

Fig. 13: Circle-skeletons with lines from the axes

OX

and

OY

(a), and marked positions of specific

points (b)

This procedure is applied for each component in

which specific points were not found by the thinning

algorithm. Why should it be done? The circle-

skeleton of the reference image of the printed circuit

board is placed in the middle of the component and

the coordinates of their pixels with a high

probability coincide with the corresponding pixels

on the manufactured image of the printed circuit

board. Thus, their respective components will be

highlighted and separated. This procedure is

required only when the board contains contacts

without traces and their defects are unacceptable.

After finding the coordinates of endings and

switches in the skeleton image the following step is

to bind chains in the reference PCB image with

these specific points. It is realized by applying the

flood-filling algorithm. Arbitrary ending or switch

pixels are taken as start points. The resulting PCB

image for filled three chains is shown in Figure 14.

All other chains are not flood-filled.

Fig. 14: The reference PCB image after flood-filling

of three chains

An order in which chains are flood-filled gives

us their ID numbers After flood-filling all endings

and switches are with corresponding colors and ID

numbers of chains to which they belong. Thus, a set

WSEAS TRANSACTIONS on CIRCUITS and SYSTEMS

DOI: 10.37394/23201.2023.22.9

Roman Melnyk, Roman Kvit

E-ISSN: 2224-266X

68

Volume 22, 2023

of pixels as specific points for the reference PCB

image

12

{ , ,..., }

n

E E E E

is formed, where

i

E

is a

set of pixels belonging to the

i

-th chain. Every pixel

is described by four parameters

{ , , , }

ij i j j

p i c x y

,

where

i

is the number of a chain,

j

is the number

of a pixel in the set

i

E

,

jj

xy

are the coordinates of

a pixel,

i

c

is the color of the

i

-th chain. Every set

i

E

contains two and more pixels. The set

i

E

also

includes pixels found for isolated contacts through

their skeletons.

4.3 Separation of Chains and Connectivity

Defects Detection

All endings and switches are placed in the middle of

contacts and traces. So, their positions are not very

sensitive to the manufacturing process. And that is

why they are accepted as the starting points to

provide manipulations with objects in the

manufactured PCB image.

As the first step a procedure to separate chains

and determine open and short defects is considered.

1. Starting points are formed by logical

projecting of the set

e

E

on the board of the

manufactured PCB. Physical operations are not

performed. Logical assignment operations are

performed

1 2 1 2

: , { , ,..., } :{ , ,..., }

nn

D E D D D E E E

,

where

D

is a set of specific points for the

manufactured PCB image and

i

D

is a set of pixels

for the

i

-th chain.

2. The filling algorithm is applied as many times

as there are chains in the scheme. Taking elements

from the set

i

D

as starting points the flood-fill

algorithm fills chains with different colors selecting

and separating them in the manufactured PCB

image.

An example of the results of the filling algorithm

is shown in Figure 15. Figure 15a shows two sets of

starting points

1{ , , , }D B C D H

2{ , , , }D A E F G

and two filled chains in the reference image. Figure

15b shows one correct chain and one chain with an

open defect. Figure 15c shows two chains as one

chain caused by a short defect. Two last examples

refer to the manufactured PCB image.

3. Defects detection. To select the red chain with

the open defect two iterations of flooding are used:

for point

B

and point

H

. Two iterations indicate an

available open defect. One color for two and more

sets of starting points indicates an available short

defect. Two blue chains with a short defect are

flooded and selected from one starting point and one

iteration.

a b

c

Fig. 15: Illustration of flood-filling and selection of

chains: chains in the reference (a), open (b), short

(c)

The resulting PCB image after three steps of

flood-filling is shown in Figure 16. One step

requires two iterations because the red chain has one

open defect resulting in two components. In the

other two steps, blue components are built from two

chains having short defects between themselves.

Fig. 16: The manufactured PCB image after three

steps (four iterations) of flood-filling

When chains are selected, and separated they are

grouped into three classes: correct, with an open

defect, and with a short defect. The coordinates of

the defects are not known. They are only visible for

external inspection. To find their positions, two

ways can be used: by application of the thinning

algorithm to the manufactured PCB image or by

comparison of two components from the reference

and manufactured PCB images. These approaches

are considered below.

4.4 Comparison of Mask Images

The binary presentation of the manufactured PCB

image is taken for the thinning algorithm. In the

WSEAS TRANSACTIONS on CIRCUITS and SYSTEMS

DOI: 10.37394/23201.2023.22.9

Roman Melnyk, Roman Kvit

E-ISSN: 2224-266X

69

Volume 22, 2023

result, there are two sets of specific points: the first

reference skeleton

12

{ , ,..., }

n

E E E E

and the

second skeleton

12

{ , ,..., }

n

D D D D

. Every element

of these sets contains two and more pixels with their

identification numbers and coordinates of detected

specific points, for example, the first chain contains

k

pixels

1 11 12 1

{ , ,..., }

k

E p p p

,

{ , , , }

ij i j j

p i c x y

.

On the basis of these two sets, two so-called

mask images are built: a mask

С

of the Etalon PCB

and

d

M

the manufactured PCB. Every mask image

has the dimension of the corresponding PCB image.

Black circles or quadrates (in our case for switches

blue and for endings red) are drawn on a white or

transparent background. Centers of circles or

quadrates coincide with the coordinates

,

jj

xy

of

specific points. A radius of a circle or a quadrate

reflects manufacturing precision. Its value is

controlled by the user. Two examples of circles and

quadrates on skeletons and not filled are shown in

Figure 17.

a

b

Fig. 17: Circles (a) and quadrates (b) marking

specific points on skeletons

Then the circles and quadrates are separated

from the skeletons and filled with the corresponding

colors as it is shown in Figure 18. Thus, the mask

image is built and prepared for the following

manipulations.

Fig. 18: Flood-filled circles

Simple observation and comparison of the mask

images do not answer how much they differ. The

special comparison procedure (1) is used to compare

two images.

The etalon mask is the output image. A pixel of

this image changes its color if two corresponding

pixels are with different intensities. A pixel becomes

red or blue if an absolute value of intensity

difference is greater than a tolerance value. Red for

a positive difference and blue for a negative one. If

the difference is within a tolerance value, the

resulting pixel remains as it was before a

comparison. For example, the separated chain with

defects is shown in Figure19a. Its mask image with

blue and red circles is shown in Figure 19b. The

reference chain and its mask are not presented. Two

mask images are accepted as operands in the

comparison formula. The resulting image from the

comparison operations is shown in Figure 19c.

a b

c d

Fig. 19: Chain with defects (a), and its mask images:

a full mask (b), difference (c), and with only defects

(d)

Blue correct circles are remained for the

illustration but are eliminated for the second

illustration in Figure 19d. Red circles mark defects

in the chain which are absent in the reference chain.

Also, some red pixels mark inaccuracies in the

positions of specific points. The reason is that the

centers of circles on the etalon PCB and the real

PCB masks do not coincide exactly. A distance

between centers of corresponding circles on the

etalon and real masks indicates a shift of the

WSEAS TRANSACTIONS on CIRCUITS and SYSTEMS

DOI: 10.37394/23201.2023.22.9

Roman Melnyk, Roman Kvit

E-ISSN: 2224-266X

70

Volume 22, 2023

concrete-specific points (corresponding contacts and

traces)

 

22

( ) [ ( ) ( )] [ ( ) ( )] ,

b b b b

D b sqrt x r x m y r y m Tol    

where b is the ID number of a specific point; r and

m are indexes of the PCB images,

Tol

and is a

tolerance value. Practically this parameter reflects

several pixels by which the real PCB image is

shifted from the etalon PCB image. It also indicates

that the real PCB is defective and must be removed.

The mask images are imposed on the original

manufactured PCB image to underline found and

suspicious defects that are demonstrated in the

example in Figure 20.

a b

Fig. 20: Overlay of the defective chain with the full

comparison image (a) and only marked defects (b)

Such visual checking is realized for every chain

having a total number of specific points greater than

the corresponding number in the reference chain.

The following approach is developed to determine

the intensity of defects and their coordinates.

5 Defects Detection by Measurement

of their Sizes

The previous methods separate every chain from the

reference and manufactured PCB image marking

places of defects and their types: short or open. The

following approach widens the opportunity for

researchers to detect other types of defects and ways

to measure their intensity.

For defects visualization, the comparison

formula is used for every chain. For the input

images as the defected chain and reference chain,

the comparison image is with positive (red) and

negative (blue) defects and shown in Figure 21a. If

required, defects can be separated by segmentation

or flood-filling of the black body of the chain

(Figure 21b). Defects have their concrete

coordinates connected with the considered chain.

a b

Fig. 21: Comparison image of two chains (a) and

their segmented version (b)

To decide on whether defects are critical for

circuit functionality their intensity must be

measured. After measurement, they are compared

with the corresponding characteristics of the

reference sample.

Surface histograms of pixels that form the

previous chain or defects are calculated by the

following formula:

( ) {( , )| ( , ) ( )}, 0,1,..., ,h i card i v I i v I black i H  

where H is the height of the image,

),( viI

is the

pixel intensity,

()hi

is the number of pixels, and

𝐼(𝑏𝑙𝑎𝑐𝑘) 𝑖𝑠 intensity of black pixels.

Their graphs are shown in Figure 22.

a b

Fig. 22:

OY

Histograms of two chains (a) and of

defect pixels (b)

Figure 22a shows histograms

OY

of the

defective (red) and reference (blue) chains. Two

marked areas indicate a difference between the

numbers of pixels in places of defects. Extra pixels

correspond to higher values; a wire break

corresponds to zero value. The second chart in

Figure 22b reflects blue and red pixels of defects in

the image received after comparison.

In this way, all inaccuracies in the components of

PCB can be measured for deciding whether the

circuit is ready for exploitation or not. Two other

examples are shown in Figure 23. The graphs

demonstrate extra and lack of pixels and thinner

traces in the defective component.

WSEAS TRANSACTIONS on CIRCUITS and SYSTEMS

DOI: 10.37394/23201.2023.22.9

Roman Melnyk, Roman Kvit

E-ISSN: 2224-266X

71

Volume 22, 2023

a

b

Fig. 23: Two components (a) and their

OX

histograms (b) with marked defect regions

The marked area of the difference between the

two graphs is measured automatically. The

difference can be caused not only by defects but not

equal wides of traces, shifts of contacts, and traces.

So, the user or special program system must take a

decision.

Defects of connectivity are detected by two

previous approaches: during the separation of a

chain having these types of defects and by

comparison of mask images. Visualizing short

defects also is useful in the approach of comparison

of reference and defective chains.

Figure 24 shows three cases: the comparison

image of two chains (with a short and a reference),

their skeleton with marked specific points, and the

overlay of two previous images. In the first image,

the blue part consists of one chain and the shorting

line.

a b c

Fig. 24: Defective chains (a), skeleton (b), and

overlay of them (c)

Measurement of defects intensity, visualization,

and determination of their coordinates helps to

divide printed circuit boards into working, defective,

and subject to repair.

6 Separation of Defects of Extra Metal

The previous approach works with all chains

defined in the control image. In the picture of the

printed circuit board, separates each circuit on the

board and specifies it as one of four possible types:

correct and the same as in the control circuit; chains

of an irregular shape or with a defective metal

filling, but conduct signals; chains with a break

defect; short circuits.

For further analysis of the image of the printed

circuit board after separation, identification of

possible defects, and classification, the circuit as an

object of consideration is removed from the

produced image. This operation is very simple: one

of the contact pixels is taken as the starting point for

the filling algorithm. The target color is white. After

this operation, the manufactured PCB image will be

fully white or with black inclusions of extra metal

on the background. In Figure 25a all checked

components instead of removed are marked with

pink for better presentation. Black extra metal traces

remained in the image. For measuring their intensity

access to them is realized through their specific

points in Figure 25b.

a b

Fig. 25: Extra metal and removed chains (a) and

specific points (b)

Measurement of these residues is required by the

user to decide the robustness of the PCB.

7 Conclusion

This approach allows us automatically to inspect all

open and short defects visible to a camera and

unnoticeable to the user. The K-means clustering

algorithm, flood-filling, thinning, and distributed

cumulative histograms are used to transform PCB

images to a binary form with uniform distribution of

colors, to find the coordinates of points identifying

every chain. The algorithms select and separate

every chain in the reference and real images, detect

WSEAS TRANSACTIONS on CIRCUITS and SYSTEMS

DOI: 10.37394/23201.2023.22.9

Roman Melnyk, Roman Kvit

E-ISSN: 2224-266X

72

Volume 22, 2023

defects of connectivity during the separation of

chains, measure the intensity of defects for chains

with extra and lack of metal filling, and visualize all

defects for every chain or the whole image.

A visualization approach is based on different

types of subtraction and comparison operations.

Overlaying images are used to project-specific

points on defective chains. It helps to indicate places

the wire breaks, short circuits, and other types of

defects.

A large number of algorithms are developed as

participants in the process of the detection of

defects. To build a robust program system of printed

circuit board inspection by their images they must

be united in four modules: preprocessing module,

open and short defects detection module, the module

for the detection of defective traces, and the module

to decide the robustness of the circuit.

References:

[1] Jungsuk Kim, Jungbeom Ko, Hojong Choi

and Hyunchul Kim, Printed Circuit Board

Defect Detection Using Deep Learning via A

Skip-Connected Convolutional Autoencoder,

Sensors, No.21(15), 2021, 4968.

[2] S. McClure. Extracting and Classifying

Circuit Board Defects using Image Processing

and Deep Learning, towardsdatascience.com,

Feb. 4, 2020 [Online]. Available:

https://towardsdatascience.com/building-an-

end-to-end-deep-learning-defect-classifier-

application-for-printed-circuit-board-pcb-

6361b3a76232.

[3] V. A. Adibhatla, H.-C. Chih, C.-C. Hsu, J.

Cheng, M. F. Abbod, and J.-S. Shieh, Defect

Detection in Printed Circuit Boards Using

You-Only-Look-Once Convolutional Neural

Networks, Electronics, Vol.9, No.9, 2020, pp.

1547.

[4] Vikas Chaudhary, Ishan R. Dave and Kishor

P. Upla, S. V., Visual Inspection of Printed

Circuit Board for Defect Detection and

Classification. International Conference on

Wireless Communications, Signal Processing

and Networking (WiSPNET), 2017, pp. 732-

737.

[5] Y. Hanlin and W. Jun, Automatic Detection

Method of Circuit Boards Defect Based on

Partition Enhanced Matching, Information

Technology Journal, Vol.12, No.11, 2013, pp.

2256-2260.

[6] Zhu, A. Wu and X. Liu, Printed circuit board

defect visual detection based on wavelet

denoising, IOP Conference Series: Materials

Science and Engineering, Vol. 392, 2018, pp.

062055.

[7] M. H.Tatibana, R. and de A. Lotufo. Novel

Automatic PCB Inspection Technique Based

on Connectivity, Proceedings X Brazilian

Symposium on Computer Graphics and Image

Processing, 1997, pp. 187-194.

[8] M. Moganti, F. Ercal, C. H. Dagli, and S.

Tzumekawa, Automatic PCB inspection

algorithms: A review, Computer Vision and

Image Understanding, Vol.63, No2, 1996, pp.

287-313.

[9] D.B. Anitha, and M. Rao, A survey on Defect

Detection in Bare PCB and Assembled PCB

using Image Processing Techniques,

International Conference on Wireless

Communications, Signal Processing and

Networking, 2017, pp. 39-43.

[10] K. P. Anoop, N.S. Sarath and V. V. Sasi

Kumar, Review of PCB Defect Detection

Using Image Processing, International Journal

of Engineering and Innovative technology

(IJEIT) Vol.4, Is.11, 2015, ISSN: 2277-3754.

[11] J. Nayaka, K. Anitha, B.D. Parameshachari,

R. Banud and P. Rashmi, PCB Fault Detection

Using Image Processing, IOP Conference

Series: Materials Science and Engineering,

Vol.225, 2017, pp. 1-5.

[12] Y. Hanlin, W. Jun, Automatic Detection

Method of Circuit Boards Defect Based on

Partition Enhanced Matching, Information

Technology Journal, Vol.12(11), 2013, pp.

2256-2260.

[13] F. B. Nadaf and V. S. Kolkure, Detection of

Bare PCB Defects by using Morphology

Technique, Morphology Technique

International Journal of Electronics and

Communication Engineering. Vol.9, No.1,

2016, pp. 63-76.

[14] S. Guan, F. Guo, A New Image Enhancement

Algorithm for for PCB Defect Detection,

International Conference Intelligence Science

and Information Engineering (ISIE), 2011, pp.

454-456.

[15] J. P. R. Nayak, K. Anitha, B. D.

Parameshachari, R. Banu, and P. Rashmi,

PCB Fault Detection Using Image Processing,

IOP Conference Series: Materials Science and

Engineering, Vol. 225, 2017, pp. 012244.R.

Melnyk, D. Hatsosh, and Y. Levus, Contacts

detection in PCB image by thinning,

clustering, and flood-filling, IEEE 16th

International conference CSIT 2021, 2021, pp.

370-374.

WSEAS TRANSACTIONS on CIRCUITS and SYSTEMS

DOI: 10.37394/23201.2023.22.9

Roman Melnyk, Roman Kvit

E-ISSN: 2224-266X

73

Volume 22, 2023

[16] W. Huang and P. Wei, A PCB dataset for

defects detection and classification. arXiv

preprint arXiv:1901.08204, 2019.

[17] S. Miles, “Computer Chip Showing High

Tech And Circuit,” 123RF.com. [Online].

Available:

https://www.123rf.com/photo_40192468_com

puter-chip-showing-high-tech-and-

circuit.html.

[18] “Circuit Board Drawing,” Circuit Board

Drawing at Painting Valley.com, Explore a

collection of Circuit Board Drawing.

Available [Online]:

https://paintingvalley.com/circuit-board-

drawing.

Contribution of Individual Authors to the

Creation of a Scientific Article (Ghostwriting

Policy)

The authors equally contributed to the present

research, at all stages from the formulation of the

problem to the final findings and solution.

Sources of Funding for Research Presented in a

Scientific Article or Scientific Article Itself

No funding was received for conducting this study.

Conflict of Interest

The authors have no conflicts of interest to declare.

Creative Commons Attribution License 4.0

(Attribution 4.0 International, CC BY 4.0)

This article is published under the terms of the

Creative Commons Attribution License 4.0

https://creativecommons.org/licenses/by/4.0/deed.en

_US

WSEAS TRANSACTIONS on CIRCUITS and SYSTEMS

DOI: 10.37394/23201.2023.22.9

Roman Melnyk, Roman Kvit

E-ISSN: 2224-266X

74

Volume 22, 2023