Você está na página 1de 5

2016 3rd International Conference on Signal Processing and Integrated Networks (SPIN)

An Efficient Image Processing Based Method for

Gills Segmentation from a Digital Fish Image
Ashish Issac1, Malay Kishore Dutta1

Biplab Sarkar2 & Radim Burget3


Department of Electronics & Communication Engineering,

Amity University, Uttar Pradesh, India
Email: issac017@gmail.com,

National Institute Abiotic Stress Management, Baramati,

Pune, Maharashtra, India
Faculty of Electrical Engineering, Burno University of
Technology, Czech Republic
Email: biplab_puru@yahoo.co.in, burgetrm@feec.vutbr.cz
deteriorate its quality, but also pose a threat to human health

Abstract The quality and freshness of a fish sample is

mainly affected due to the handling and storage conditions
during the post harvesting period. The retention time and storage
medium are the two main factors affecting the fish quality. This
paper presents an image processing based method for automatic
and efficient segmentation of gills from the fish sample image
which can be used for fish freshness validation and determination
of any pesticide from the fish sample under test. The
implemented algorithm has produced a maximum correlation of
92.4% with the ground truth results obtained from experts. The
method used for gills segmentation is fast and simple.

Fish freshness can be judged using a number of parameters

such as skin appearance, gills color, firmness and elasticity of
the fish and odour of the fish gill [4]. The visually
distinguishable features which can be observed by human eye
can be used for image processing based analysis of the fish
samples. Of the above mentioned parameters, gill colour is the
suitable parameter for image processing analysis. The gill
colour is red coloured respiratory tissue of the fish. The gill
colour along with the odour has been used as a manual
freshness indicator by fishermen and customers till date [6].

KeywordsImage Processing, Fish, Gills, Image Segmentation.

Aquaculture has emerged as one of those food producing
segments that is growing fast along with field crops and
livestock productivity [1]. The atmospheric catastrophe and
pollution impacts have been a major cause for the decline in
the quality of the captured fisheries. An increase in the fish
production does not imply that the captured fish is healthy and
safe for human consumption. Since fish is a perishable entity,
indicators such as freshness and quality are important for the
commercial success.

Image analysis has several advantages when compared to

subjective freshness indicator. It is a non-destructive, nonhazardous, objective and relatively cheap tool for evaluating
the quality of fish [7]. Some work has already been reported in
food processing sectors using the image processing. Feng
Wang et al [8] proposed a method for fish freshness
determination from the eye using a regression based
technique. Fairuz Muhamad et al [9] revealed a method for
classification of fish freshness using fuzzy logic. Shiv Ram
Dubey et al [10] described a method to detect and segment the
infected region in a fruit image using a colour based
technique. Soumya et al [11] used soft computing techniques
such as c-means clustering and fuzzy logic for segmentation of
color images. Shiv Ram Dubey et al [12] has used histogram
analysis of the texture based features to detect defects in fruit
images. Although all the above methods have been used for
segmentation and analysis of the segmented images, yet there
is a need for the development of a faster and simpler algorithm
which can be used in real time for the segmentation of the
objects from the color images.

The fishes which are harvested at local ponds are not

always consumed by nearby people. However, they are being
transported to some hot spots, where fishes are popular for
profitability and marketability. One of the best methods for
preservation of the fish decaying process and prime insulator
during transportation is ice preservation [2]. However, the
preservation efficiency depends on fish origin, type of fish,
climatic conditions and quality of ice used for preservation.
The storage temperature and post harvesting time, between
death and consumption, are important factors for fish quality
[2]. Temperature can be a cause of autolytic and bacterial

The main contribution of this paper is a novel image

processing based method for gills segmentation which can be
used for validation of fish freshness and presence of any
pesticides in a fish. The image processing is a non-destructive
technique and does not affect the quality of the sample. For
this purpose an adaptive threshold based method has been
introduced in this paper, which is fully automatic, runs in realtime, cheap for implementation, computationally efficient and

The post harvested fishes have posed multiple cases of

compromised quality with respect to the series of parameters
related to safety, nutrition, freshness and edibility [3]. The loss
of food characteristics in terms of taste and quality may result
due to physical and biochemical changes in fishes during post
harvesting period [4]. Selling of infected fishes can not only

978-1-4673-9197-9/16/$31.00 2016 IEEE


2016 3rd International Conference on Signal Processing and Integrated Networks (SPIN)

hazardous tools to evaluate photography based data. So, the

images of gills from post harvested fish samples were captured
at regular time interval.

ready for embedding to hardware device. The proposed

method resulted in the fish gills segmentation in 3.45 seconds.
The proposed method was compared to human expert results
and has achieved correlation of 92.4%.

The flowchart of the proposed methodology is shown

below in Fig 2. The fish gills segmentation from a fish sample
image can be divided into 3 basic blocks. The first block is the
conversion of the input sample image into a suitable colour
space which is best for the segmentation. The automatic
threshold is applied to the colour space transformed image in
order to segment the gills accurately and efficiently. The final
block is the correlation between the manually segmented
ground truth marked by the experts and the automatic
segmented gills obtained as an output of the software code.

The remaining paper is divided into 3 sections. Section II

discusses about the materials and methods which have been
used for the experiment. Section III discusses about the results
and Section IV is related to the conclusions from the results
and some discussion.
The fish samples which are used for experimentation
purpose consists of various body parts in the sample image.
Out of the many body parts, the region of interest for the
proposed work is the fish gills. The gills are a red coloured
respiratory tissue of fish and the best suited organ for
freshness and quality determination of a fish [4].

Colour Space

Extracting a
channel from
LAB colour
space image

Colour Space Conversion



Fig.1: Labeled fish sample

Fig. 1 show a fish sample with different parts of the sample

marked on it. The fish mouth is a conical or cylindrical shaped
opening which serves for taking in food. The nostrils are
directly connected with the organ of sense, which helps the
fish in identifying the food at some distance. The eye of the
fish has round pupils which do not vary in size. The
operculum is a hard bony structure which protects the gills. It
is also called gill cover and is in the form of a flap. The gills
are useful in absorbing the oxygen from the water. The mouth
floor and operculum helps to produce a stream of water into
the mouth, over the gills and out from the operculum. The gill
rakers are bony structures which are used to divert the prey to
the esophagus and not letting them enter the gills filament.
The gill filaments are the fleshy part of the gills which are soft
and red in color. The gill filament increases the surface area to
help the fish in absorbing the oxygen from water. The gill
filaments are the smallest subdivision of the gills. The pectoral
fins are situated just behind the operculum on both sides. The
pectoral fins helps in maintaining the lifting force to help fish
to maintain depth. The fish scales are plate like structure
which covers the skin of the fish. The scales can help in
identification of species of a fish.

Fish Gills

Ground Truth
and Segmented

ROI Segmentation and Correlation

Fig. 2: Flowchart of proposed methodology

The captured fish sample image is an RGB image. On

analyzing all the 3 RGB channels, i.e. red, green and blue, of
the input image, it was observed that the gills segmentation is
not accurate as it contained a lot of noise in the form of eye
and scales. It was possible to use segmented results, however
this would request additional computational power. Instead of
this the image represented in RGB colour space was converted
into different colour spaces and on analyzing it was observed
that the LAB colour space and a component was best suited
for gills segmentation.
The a channel is thresholded on the basis of the intensity
using Otsu Threshold method to convert the image to a binary
image. The thresholded binary image consists of fish gills and
eyes and fins. The fish eyes and fins are considered as noise
and removed from the binary image using area thresholding.

The gill colour and odour has been used as a parameter for
fish freshness by fishermen and customers till date. Also,
image processing techniques are non destructive and non


2016 3rd International Conference on Signal Processing and Integrated Networks (SPIN)

As is clear from the fig. 1, the gills have largest area among
the eyes and fins, so the pixels in binary image are grouped on
basis of area and then the labeled region of maximum area is
segmented as gills.

used to segment the gills and eyes from the image, which can
be seen in Fig 5(c). Some portion of the eye, which is
extracted along with the gills is treated as noise and needs to
be eliminated. It can be seen that the gills is the region of
largest area in the segmented binary image. So, an area based
threshold is applied to the binary image to segment the gills.

The gills in the input image are manually marked by

experts and the marked gills are termed as ground truth. A
correlation is established between the gills which are
segmented using the code and the ground truth to measure the
accuracy of the code.
A total of four fishes were used for the experimentation
purpose. All the fishes were taken from the fish ponds at
National Institute of Abiotic Stress Management, Pune. The
average length of the fish samples is 21.600.50 cms. The
average weight of the samples is 90.401.20 gms. The fishes
were taken out from the pond and were placed in ice chilled
water for sudden death. This sudden death of the fish prevents
the sample from rigor mortise. Once the sample was lifeless,
the imaging was done for the next 13 days with a two days
interval between successive images. The camera used for
image acquisition was NIKON D90 and the size of the
captured images was 601 301 pixels.

Fig. 4: LAB colour image (a) LAB (b) L channel (c) a channel (d) b

Fig. 3 demonstrates the selection of the best suitable colour

space for gills segmentation. Out of the many colour spaces,
the input RGB image of the fish sample was converted into
LAB colour space. It is observed from Fig 3(b), that the gills
are clearly distinguished by green colour (i.e. value of a
channel) from the remaining fish body. So, LAB colour space
and channel a is used for the segmentation.

Fig.5: Gills segmentation (a) Input RGB image (b) 'a' channel (c) intensity
based threshold (d) area based threshold

Fig. 6 shows the result of gills segmentation for a fish

sample on a particular day. Fig 6(b) shows the gills segmented
from the fish image using the software code while Fig 6(c) is
the ground truth segmented gills which is marked by the
experts. A correlation between both the code and ground truth
segmented gills images is established and the results are
tabulated below in Table I.

Fig. 3: Colour space conversion (a) Input RGB image (b) LAB colour space

Fig. 4 shows the LAB colour space transformed image and

its constituting channels. From Fig 4(c), it can be observed
that the fish gills are easily distinguished from the
background. Fig 4(c) is the a channel and it is a
representation of the position of a particular colour between
red and green. The gills, being of red colour in RGB image
and green in LAB colour image, are easily distinguished from
other objects in the fish image.
Fig. 5 demonstrates the segmentation of the gills from the
a channel of LAB image. As it can be seen from Fig 5(b),
that there is a great contrast between the gills and the eyes and
the other objects, so an automated intensity based threshold is

Fig.6: Correlation (a) Input RGB image (b) Code segmented fish gills (c)
Ground truth segmented fish gills


2016 3rd International Conference on Signal Processing and Integrated Networks (SPIN)


The correlation coefficients for each fish sample is

tabulated above and represented graphically in Fig 7. The xaxis represents the day on which the data was collected and yaxis represents the correlation coefficient between the gills
segmented and the ground truth. The range of correlation
coefficient lies in between 0 and 1. It can be observed from the
figure that the correlation coefficient has been greater than 0.8
for most of the images. The maximum correlation achieved is



Correlation between
Ground Truth and
Code Segmented Gills

Day 1


Day 2


Day 3


Day 4


Day 5


Day 6


Day 1


Day 2


Day 3









The fish gills are one of the many visual parameters which
can be used to determine the freshness and quality of a fish
sample. Also, the fish gills can be used to test if the fish is
polluted by any pesticides. In this paper we introduced a novel
image processing based method for gills segmentation which
can be used for validation of fish freshness and presence of
pesticides in a fish. The image processing approach has
advantage that it is a non-destructive technique, is cheap, fast
and does not affect the quality of the sample. The maximum
correlation between the code segmented and the human expert
segmented fish gills for the samples under test is 92.4%.
However, this value can be increased if care is taken while
manual uncovering of the gills. Special care should be taken
so that the blood does not spill on other body parts as this may
lead to error in the gills segmentation using image processing.

Day 4



Day 5



Day 6



Day 1



Day 2


Day 3


Day 4




Day 5


[1] Troell M, Naylor RL, Metian M, Beveridge M, Tyedmers PHet


Day 6



Day 1



Day 2


Day 3


Day 4



Day 5



Day 6


al(2014), Does aquaculture add resilience to the global food

system?,Proc Natl AcadSci U S A. 16;111(37):13257-63.
Rahman MM, Nasrun M, Hossain MY, Aa'zamuddin M(2012),
A comparison between ice and salt storages on bacteriological
quality of Asian seabass (Latescalcarifer), Pak J Biol Sci. 2012
Jun 15;15(12):583-8.
Hossain MM, Rahman M, Hassan MN, Nowsad AA(2013),
Post-harvest loss of farm raised Indian and Chinese major carps
in the distribution channel from Mymensingh to Rangpur of
Bangladesh,Pak J Biol Sci. 2013 Jun 15;16(12):564-9.
G. Olafsdottir, E. Martinsdottir, J. Oehlenchlager, P. Dalgaard,
B.Jensen, I. Undeland, I. Mackie, G. Henehan, H.
Nielsen(1997), Multi-sensors for fish quality determination,
Trends Food Sci. Technol. 8, 258265.
Shatabdi Goon, MunmunShabnamBipasha, Md. Saiful Islam
(2014), Fish marketing status with formalin treatment in
Bangladesh. Online Journal of Social Sciences ResearchISSN
2277-0844, 49-53.
A. Macagnano, M. Careche, A. Herrero , R. Paolesse , E.
Martinelli ,G. Pennazza , P. Carmonae, A. DAmico, C. Di
Natale (2005), A model to predict fish quality
instrumental features, Sensors and Actuators B 111112 , 293
Menesatti, P., Costa, C., Aguzzi, J. 2010, Quality evaluation of
fish by hyperspectral imaging, Hyperspectral Imaging for Food
Quality Analysis and Control, Chapter 8, pp. 273-294,
Academic Press / Elsevier, San Diego, California, USA.
Feng Wang, Yue Zang, Qiqi Wo, Chan Zou, Nan Wang, et al,
Fish Freshness rapid detection based on fish eye image, Proc.
SPIE 8761, PIAGENG 2013: Image Processing and Photonics
for Agricultural Engineering, 87610A (March 4, 2013).











Fig 7 Graph showing correlation coefficient between segmented and ground
truth fish gills


2016 3rd International Conference on Signal Processing and Integrated Networks (SPIN)

[9] Muhamad F, Hashim H, Jarmin R, Ahmad A, Fish freshness

segmentation technique International Journal of Artificial

Intelligence and Interactive Multimedia, 2013 Vol. 2, N 2.
[11] Sowmya Band, B. Sheelarani, Colour image segmentation
using soft computing techniques, International Journal of Soft
Computing Applications, vol. 4, pp. 69-80, 2009.
[12] Shiv Ram Dubey, Anand Singh Jalal, Fruit disease recognition
using improved sum and difference histogram from images,
International Journal of Applied Pattern Recognition, Vol 1,
Issue 2, Pg 199-220, 2014.

classification based on image processing and fuzzy logic,

Proceedings of the 8th WSEAS International Conference on
Circuits, Systems, Electronics, Control &; Signal Processing,
World Scientific and Engineering Academy and Society
(WSEAS), Stevens Point, Wisconsin, USA, CSECS09, pp 109
[10] Shiv Ram Dubey, Pushkar Dixit, Nishant Singh, Jay Prakash
Gupta Infected fruit part detection using K-means clustering