You are on page 1of 6

INTERNATIONAL JOURNAL OF INNOVATIVE RESEARCH EXPLORER ISSN NO: 2347-6060

A Framework for Underwater Image


Enhancement and Object Detection
1
Thamarai M , 2Dhivyaa S P
1
Professor, Department of ECE, Malla Reddy College of Engineering, Hyderabad, India.
1
mthamarai2014@gmail.com
2
System Hardware Associate, Google, Hyderabad
2
spdhivyaa@gmail.com

Abstract— the underwater images are distorted due to the light absorption and scattering. The enhancement techniques used to
improve the quality of the underwater images are based on the level of noise introduced and also the water quality. Underwater image
enhancement processing is necessary to detect desired or suspicious objects under the sea. This paper describes an improved image
enhancement and object detection for the underwater images in YCbCr space and canny edge detection for underwater images. The
edge detection is done for each Y, Cb and Cr components separately and are combined to get more accurate edge detected underwater
image. The morphological process is also applied to obtain the thin edges and the identified objects parameter like area, length, width,
etc are also calculated. The results obtained from the proposed method helps efficient object detection.

Keywords— underwater images, color segmentation, histogram, number of edges.

I. INTRODUCTION
The quality of underwater image plays a vital role in scientific missions such as monitoring sea life, taking census of populations
of fish and biological environments. Capturing images underwater is challenging, due light reflection from the water surface and
scattering of light by water particles inside the sea. color change due to varying degrees of light attenuation for different
wavelengths. Thus the acquired under water images have low contrast with color distortion. Several researchers proposed
different techniques for underwater image enhancement and detection. The underwater image processing differs from normal
image processing due to noise and poor illumination [1], [2]. Conventional techniques use color equalization for enhancing the
underwater images. Images enhanced using these techniques are not giving promising results for edge detection, desired
parameters extraction and analysis. In particular, blue and green colors are dominant in underwater images. The colors like yellow
and red almost disappear while we go increasing depth. So, efficient image enhancement and edge detection is required for
underwater analysis.

The color components are not separated in RGB model. Because of it, image enhancement and accurate object detection is not
possible. Meanwhile YCbCr model separates the input image into three components and gives flexibility in modifying images. In
this paper, the algorithm enhances the brightness, visibility and contrast of underwater images. The enhanced image is subjected
to edge detection. The proposed technique uses YCbCr model for processing. Enhancing the luminance component in this model
automatically enhances the brightness of the image without disturbing Cb and Cr components.

Object detection in underwater image is difficult since the object edges are affected due to light reflection on water surface. Also
in deep water, the boundary of the object is not detected accurately due to dark of the object and poor illumination. The enhanced
images produce an improved edge detection results. In this paper section II describes the literature survey and section III describes
the proposed work. Section IV and V explains the results with evaluation measurements and session VI concludes the paper.

II. EXISTING WORK


Edge detection is most widely used in image processing and analysis of the underwater images such as feature description image
segmentation, pattern recognition, etc. The classic edge detection methods require pre-processing for efficient underwater images.
So pre-processing is a key factor for obtaining efficient feature extraction. In [3], object detection in underwater image is
described. In that paper, first the RGB image is converted into blue image and then LOG operator is used for edge detection.

VOLUME 4, ISSUE 6, NOV/2017 118 http://ijire.org/


INTERNATIONAL JOURNAL OF INNOVATIVE RESEARCH EXPLORER ISSN NO: 2347-6060

The advantages of the LOG operator over the other operators are the edge smoothening by Gaussian function and the rational
invariance by laplacian portion. A disadvantage of the operator is the large size of the kernels.

In [4], enhancement of underwater images using CLAHE Algorithm is explained. RGB color space is converted into L*a*b color
space and CLAHE algorithm is applied to L only. Yuejiao et la [5] used K-means algorithm for underwater image object detection
and classification.

Banerjee [6] proposed a real time underwater image enhancement technique for imaging with AUV150 images. In that paper [6],
sequence of linear, non-linear filters and contrast correction in the RGB and YCbCr color space is used to enhance the underwater
images. This multi-step image enhancement algorithm RyPro is suitable for real time-applications.

In [7], an automated approach for fish detection was proposed. The shape information of the fish is incorporated into level sets
through PCA method.

PSO algorithm is used to enhance color image. In this [8], first RGB values are adjusted using PSO and enhancement using PSO
was done in HSV color space. These techniques are time consuming and have more computational complexity. The proposed
method is simple and more suitable for real-time applications such as fish monitoring and suspicious object detection in
underwater.

III. PROPOSED WORK


The proposed method consists of two steps which involves pre-processing of the acquired image followed by the edge detection.

Fig.1 Pre Processing Block Diagram

The pre-processing consists of the following steps and is shown in figure1..


i) Imfilter
ii) Histogram Equalization
iii) Colour Space Conversion
iv) Edge Detection

Imfilter is used for multidimensional image filtering. It filters the multi-dimensional array using multi-dimensions filters. The array
can be a non-sparse numeric array of any class and dimension. The resultant array is of same class and size of the input array.

Histogram Equalization is the process of automatically adjusting intensity values. It involves changing the intensity values so that
the histogram of the output image approximately matches the desired histogram. The transformation helps in enhancing the contrast
of the image. It can be achieved by converting an RGB image to an YCbCr image and enhancing the intensity.

Color Space Conversion is a process of converting the RGB color space into its equivalent components and colorimetric values
depending upon the application we are using. Here YcbCr color space process is used. YcbCr color space is widely used in digital
video. Since the image is acquired by a camera, we are using this Color space. The video obtained from the camera is split into
several frames and every tenth frame is taken for the image processing. In this format, the luminance information is stored as a
single component (Y) while the chrominance components are stored in two color difference components (Cb and Cr). However,
performing histogram equalization on R, G and B components separately will not enhance the image. The block diagram for edge
detection is shown in figure 2.

Edge Detection is an image processing technique used to find the boundaries of an object in an image. Edges normally shows the
sudden variation in pixel intensity values in an image. Edge detection technique works by detecting discontinuities in terms of
brightness. Edge detection is used in data extraction and segmentation. Common edge detection algorithms include Roberts,
Perwitt Sobel, Canny, Roberts and fuzzy methods. Most of the information about an object in an image is defined by the edges.

VOLUME 4, ISSUE 6, NOV/2017 119 http://ijire.org/


INTERNATIONAL JOURNAL OF INNOVATIVE RESEARCH EXPLORER ISSN NO: 2347-6060

Fig. 2 Edge Detection Block Diagram

Here we are using the canny operator for detecting the edges of the given image because of its better performance in noisy
environments. Canny edge detection is a technique to extract structural information from different objects in an image and
reduces the amount of data to be processed. It has been widely used in computer vision systems.

The general condition for edge detection include:

1. Detection with low error rate, means that the edge detection should accurately get as many edges for the given image as
possible
2. The edge point detected by the operator should accurately satisfy the local properties of the image.
3. Sometimes noise pixels may detected as edges pixels, so the detection techniques has to identify such a false edges and
should not be marked as edge.

Among the edge detection methods developed so far, canny edge operator provides good and reliable detection. It uses optimality
based on non maxima criteria for edge detection. Because of its process simplicity ,It is easy to implement.So it became popular
algorithm for edge detection.

Algorithm:
1. Convert the RGB color space image into YCbCr color space image.
2. Histogram equalization is applied in the obtained YCbCr image which enhances the contrast of the image.
3. Now, the enhanced image is split into its respective Y, Cb and Cr components.
4. Canny operator is applied to each of the Y, Cb and Cr component images.
5. All the three edge detected image are now combined to form a single YCbCr edge detected image.
6. The obtained YCbCr edge detected image is dilated inorder to obtain the thin edge.
7. Now, the boundary of the thin edge image is traced.
8. Finally, the boundary line indicates the detected object.

IV. RESULTS AND COMPARISON

The proposed algorithm has been implemented using MATLAB. This implemented method is tested on various underwater
images. The figure 3 shows the original and enhanced image and its respective histograms.

Fig. 3 Input and enhanced image with histogram

VOLUME 4, ISSUE 6, NOV/2017 120 http://ijire.org/


INTERNATIONAL JOURNAL OF INNOVATIVE RESEARCH EXPLORER ISSN NO: 2347-6060

The figure 4 shows the Y, Cb and Cr image after histogram equalization.

Fig. 4 Y, Cb and Cr Components

The figure 5 shows the edge detected image of Y, Cb and Cr components and the combined edge detected YCbCr image.

Fig. 5 Edges of Y, Cb, Cr Components and Combined Edges

The figure 6 shows the dilated image of the combined YCbCr image, thin edge detected image, boundary tracing and object
detected image.

Fig. 6 Image after dilation with boundary tracing

VOLUME 4, ISSUE 6, NOV/2017 121 http://ijire.org/


INTERNATIONAL JOURNAL OF INNOVATIVE RESEARCH EXPLORER ISSN NO: 2347-6060

The following figure 7 shows the image enhancement and detection of two underwater images.

Edge Detected Image Edge Detected Image

Object Detected Object Detected

Fig.7 Underwater images and object detected


The performance of the algorithm is tested using underwater fish images. The detected object (fish) size (area) and width are
measured and tabulated in table 1. And also the detected object index number is also given. The index number denotes the
maximum size of the connect component in the image.

TABLE I: Detected Object Width and Area

Detected Object Size Index


Input
Area Width number
Image 1 745 443 7
Image 2 1141 384 14
Image 3 1810 499 2

V. CONCLUSION
Underwater image enhancement and object’s edge detection has been discussed in this paper. The object extracted using this
method, produces better results than canny edge detection without enhancement. In this method, edge detection is applied
separately for Y, Cb and Cr components.

VOLUME 4, ISSUE 6, NOV/2017 122 http://ijire.org/


INTERNATIONAL JOURNAL OF INNOVATIVE RESEARCH EXPLORER ISSN NO: 2347-6060

The proposed technique has clear potential to identify fish population from the video sequences taken from noisy underwater
environment. It is also useful for earth remote sensing images. In future, restoration models can be used to improve performance
of the proposed enhancement and object detection system..

REFERENCES
[1] Iqbal K, Salam R A, Osman A and Talib A Z 2007 “Underwater image enhancement using an integrated color model” IAENG Int. J. Computer Science 32(2):
239–244

[2] Sedlazeck A and Koch R 2011 “Simulating deep sea underwater images using physical models for light attenuation, scattering and refraction” Vision,
Modeling, and Visualization
Workshop, pp 49–56

[3] Niranjan Anand Hule, “Underwater Image Processing For Object Detection”, International Journal of Innovative and Emerging Research in Engineering
,Volume 2, Issue 2, 2015

[4] V.Vembuselvi, T.Murugan, “Enhancement of Underwater Images Using CLAHE Algorithm”, International Journal of Innovative Research in Science,
Engineering and Technology, Volume 5, Special Issue 2, March 2016

[5] Yuejiao He, Hua Yang, “underwater image edge detection based on K-means algorithm”

[6] JEET BANERJEE, RANJIT RAY, “Real-time underwater image enhancement: An improved approach for imaging with AUV-150” Indian Academy of
science, Vol. 41, No. 2, February 2016, pp. 225–238

[7] C.R.Savage, R.J. Petrell, ”Underwater fish-video images: image quality and edge detection technique”, ASAE AQUA- 107, pp 31 may 1994

[8] Amal Abusnaser, Tyad Abu Doush ”Underwater image enhancement using Particle swarm optimization” Journal of Intelligent systems 2014, pp 1-18, DOI
10.1515/jisys

Dr.M. Thamarai
She has received Ph.D degree in Digital Image processing from the Anna University Chennai . She has been
working as professor and Dean in Malla Reddy College of Engineering ,Hyderabad since 2017. She has
participated and published papers in around 20 National and Internal Conferences and also published 18 papers
in National and International journals. Her research interest includes Digital image processing and video coding
and VLSI implementation of image processing algorithms.

Ms. S P Dhivyaa
She has completed Bachelor of Engineering in Bannari Amman Institute of Technology Erode Tamilnadu. She
has participated and published papers in 3 National and International Conferences. Her research interest
includes Digital Image Processing, Machine Vision and VLSI algorithms.

VOLUME 4, ISSUE 6, NOV/2017 123 http://ijire.org/

You might also like