e-ISSN: 2582-5208

International Research Journal of Modernization in Engineering Technology and Science

( Peer-Reviewed, Open Access, Fully Refereed International Journal )
Volume:04/Issue:06/June-2022 Impact Factor- 6.752 www.irjmets.com
A GENERATIVE ADVERSARI AL NETWORK BASED DEEP LEARNING
METHOD FOR LOW QUALITY DEFECT IMAGE RECONSTRUCTION
AND RECOGNITION
S. Ruchitha*1, T. Sony*2, T. Akhila*3, D. Sandhya Rani*4
*1,2,3Authors, Department Of Computer Science Engineering, CMR Technical Campus,
Kandlakoya (V), Hyderabad, Telangana, India.
*4Assistance Professor, Department Of Computer Science Engineering, CMR Technical Campus,
Kandlakoya (V),Hyderabad, Telangana, India.
ABSTRACT
Machine vision significantly improves the quality, efficiency ,and reliability of low quality detection of image. In
the field of visual inspection, vision-based defect recognition deep learning plays important role. Basically a low
quality image may lose some useful information. To eradicate this problem we use a GAN’s network. Generative
adversarial networks are thriving unsupervised machine learning techniques. And GAN’s are significant in
various fields such as natural language processing, computer vision. GAN’s used to reconstruct the low quality
image to high quality resolution image. Basically GAN’s consists of two networks generator network,
discriminator network. The work of generator network is to take input data and the input image of low quality
defect image and generate high resolution image. The work of discriminator is to recognize whether the
reconstructed image having high quality compared to the defect image. Working of both networks gives a final
image output. Methods like PSNR, SSIM ,MSE used to calculate the quality of both degraded and high resolution
image and by comparing both values we can recognize whether it is constructed properly or not.
Keywords: GAN, Generator Network, Discriminator Network, Degraded Image, High Resolution Image.
I. INTRODUCTION
Now-a-days a lot of crime scene , accidents ,and harassment are happening in society. If we have a proof in
form of image of low quality . we used SRGAN in this project. This system is used to convert that low quality
image to HR image. So by converting we can try to catch the victim and some useful information is discovered.
We offer a Deep learning expert way for single picture building manager power to get at detail . Our carefully
worked design straight-forwardly learns a start to end mapping between the low/high-resolution pictures. The
Purpose of this system is to discover the low quality image and also remake the low quality image to high
power to get at detail using a deep learning careful way GAN. The low-quality images usually do not keep some
useful news given and may mislead the DL methods into poor outcomes. To over-come this hard question, this
undertaking proposes a producing Adversarial network (GAN)-based DL careful way for low-quality bad,
wrong point image recognition. Generative Adversarial networks GAN’S are a getting on well un-overseen
machine learning expert way that has led to important moves-forward in different fields such as knowledge
processing machine act or power of seeing, natural language processing, among others .We will be using SRGAN
in our undertaking, that takes the low-detail picture as the news given and gives in the high-detail one. We
further see that as generally done not (good) enough coding-based SR techniques can in the same way be seen
as a deep conovloutional order. In any example, not the same as common carefully worked designs that grip
every part freely, our carefully worked design together free smooth-moving current at all levels. Our Deep
SRGAN has a lightweight structure, yet gives signs of best in part getting back quality, what is more, does a
quick rate of motion for able to keep living on-line use. We make observations of different system structures
and parameter frames to do trade-offs among execution and rate of motion. In addition, we stretch out our
system to adjust to three color narrow ways at the same time, and let see better over-all remaking quality.
II. METHODOLOGY
GENERATIVE ADVERSARIAL NETWORK
The proposed system makes use of generative adversarial network (GAN)-based DL method for low-quality
defect image recognition Image .Generative adversarial networks are thriving unsupervised machine learning

www.irjmets.com @International Research Journal of Modernization in Engineering, Technology and Science

[1668]
 e-ISSN: 2582-5208
International Research Journal of Modernization in Engineering Technology and Science
( Peer-Reviewed, Open Access, Fully Refereed International Journal )
Volume:04/Issue:06/June-2022 Impact Factor- 6.752 www.irjmets.com
techniques. GAN’s consists of two networks generator network , discriminator network. Super resolution has
been implemented in several different ways using Technologies like MATLAB. The system shows an acceptable
accuracy for image super resolution by implementing Neural Networks in the system. The results under
degraded images show that the proposed method achieves very better performances, which has accuracy
ranged from 95.53-99.62%. With different masks ,disturbances and noises, and they are improved greatly
compared with the other methods. Furthermore, the results on PSNR, SSIM indicate that the quality of the
reconstructed image is improved greatly, which is very helpful for defect analysis .Implementation of Machine
Learning and Neural Network increases the efficiency and Accuracy of the system.
III. ARCHITECTURE

Figure 1: Architecture
In the architecture first raw image data is taken as input .Input should be in image format. After that image
undergo under the image data acquisition .image data acquisition is a step in the image preprocessing. It is a
process to retrieving image from external source this is used for further processing. After image data
acquisition the image undergo under image preprocessing .In the image preprocessing the image is formatted
for suiting the model. This includes resizing of image etc. Later image degrading is done which means the losing
the quality of image.
This degraded image is sent to GAN model. It is basically a network consist of two neural networks. These two
networks compete with each other and produce the changes in the data. It is unsupervised machine learning
technique which consist of two networks generator network, discriminator network. The work of generator
network is to take input data and the input image of low quality defect image and generate high resolution
image. The work of discriminator is to recognize whether the reconstructed image having high quality
compared to the defect image.
Model inference allows the user to communicate with the model. Then the super resolution image is produced
as output. Then after PSNR(peak-signal-to-noise-),SSIM(structure similarity index measure),MSE(mean square
error) values are calculate

www.irjmets.com @International Research Journal of Modernization in Engineering, Technology and Science

[1669]
 e-ISSN: 2582-5208
International Research Journal of Modernization in Engineering Technology and Science
( Peer-Reviewed, Open Access, Fully Refereed International Journal )
Volume:04/Issue:06/June-2022 Impact Factor- 6.752 www.irjmets.com
IV. RESULTS AND DISCUSSION

Figure 2: HOME PAGE SCREEN

Figure 3: INPUT IMAGE RESULT

Figure 4: OUTPUT WITH DEGRADATION

www.irjmets.com @International Research Journal of Modernization in Engineering, Technology and Science
[1670]
 e-ISSN: 2582-5208
International Research Journal of Modernization in Engineering Technology and Science
( Peer-Reviewed, Open Access, Fully Refereed International Journal )
Volume:04/Issue:06/June-2022 Impact Factor- 6.752 www.irjmets.com

Figure 5: OUTPUT WITHOUT DEGRADATION

Figure 6: TEST GRAPH

V. CONCLUSION
With ongoing crimes, it is clear that the evidence is compulsory for solving the crime. If image is of low quality
then we need to get a high resolution image. In vision-based defect recognition, deep learning (DL) is a research
hotspot. However, DL is sensitive to image quality, and it is hard to collect enough high-quality defect images
Although state-of-the-art non-machine learning algorithms for image denoising exist, we are constantly
wondering if we can achieve better performance with the assistance of deep learning .This system proposes a
generative adversarial network (GAN)-based DL method for low-quality defect image recognition. Generative
Adversarial Networks, or GAN’s for short, are an approach to generative modeling using deep learning methods,
such as convolutiona l neural networks.
VI. REFERENCES
[1] Aharon, M., Elad, M., Bruckstein, A.: K-SVD: An algorithm for designing over complete dictionaries for
sparse representation. TSP 54(11), 4311–4322 (2006).
[2] Bevilacqua, M., Roumy, A., Guillemot, C., Morel, M.L.A.: Low-complexity single image super-resolution
based on nonnegative neighbor embedding. In: BMVC (2012).
[3] Burger, H.C., Schuler, C.J., Harmeling, S Image denoising Can plain neural networks compete with
BM3D? In: CVPR. pp. 2392–2399 (2012).
[4] Chang, H. Yeung, D.Y., Xiong, Y.: Super-resolution through neighbor embedding. In: CVPR (2004).
[5] Deng, J., Dong, W., Socher, R., Li, L.J., Li, K., Fei-Fei, L.: ImageNet: A large-scale hierarchical image
database. In: CVPR. pp. 248–255 (2009).

www.irjmets.com @International Research Journal of Modernization in Engineering, Technology and Science

[1671]