Automatic Image Detection and Censoring On Web Pages
Automatic Image Detection and Censoring On Web Pages
Volume 5, Issue 6, June – 2020 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
Automatic Image Detection and
Censoring on Web Pages
Adithye Joseph1, Ajay Chacko Thomas1, Jobin Jose1, Divya Sunny2
1
UG Student, Dept. of CSE, SJCET Palai, Kerala, India
2
Assistant Professor, Dept of CSE, SJCET Palai, Kerala, India
Abstract:- AIDC(Automatic Image Detection And off and access other controls. The users/children can now
Censoring) is a software that utilizes AI and Machine access the internet via the web browser on the computer.All
learning to automatically detect and censor the contents the content they access will now be automatically censored
on a web page. As we all know ,the internet is a vast by AIDC. As the control panel of the censoring system is
space with enormous amount of data and content in it password protected the users cannot make any changes to
and a large portion of these contents are pornographic the settings set by the administrator. The control panel
and violent images and videos.And the current situation gives the administrator access to various settings such as
is such that even a child can easily access these contents. excluding certain websites from being censored.The
So what AIDC aims to provide is a software that can be administrator can also turn off the censoring when
installed on the computer that will automatically censor required.The software primarily focuses on blurring of guns
these contents and thus keeping the young minds safe on to abstract the idea of violent image censoring on online
the internet.This feature can controlled by an content.Censoring is only done on necessary areas so as to
administrator.Due to limitations,the scope of this preserve the meaning of the image.This method is far
project will include only censoring of violent images. superior to existing methods which completely blurs or
hides the image destroying its intended purpose.
I. INTRODUCTION
II. PROPOSED METHOD
The easy access and widespread of the Internet makes
it easier than ever to reach content of any kind at any The working of the entire process is as
moment, and while that poses several advantages, there is follows:Webscraper extracts images from the web page.The
also the fact that sensitive audiences may be inadvertently images are then stored in a folder locally.An image
exposed to violent content they did not ask for.AIDC aims detection model then access this folder to retrieve images
to provide is a software that can be installed on the from it.The images are then processed by the model to
computer that will automatically censor these contents and detect the relevant object.A masking algorithm generates a
thus keeping the young minds safe on the internet.After mask for the object concurrently.The mask is then placed
installing the AIDC plug-in/software the admin of the on the object to generate the censored image.The image is
system or the parent can access the AIDC control panel then send back to be viewed bytheuser.
which allows him/her to turn the censoring system on and
Chart 1:- Overall Design
The entire project can be divided into three sections. B. Detection of Objects
Here object detection is done using Mask R-
A. Extraction of Images CNN.Mask RCNN is a deep neural network aimed to solve
A custom made webscraper is made to extract images instance segmentation problem in machine learning or
from webpages. A web crawler is used to browses the computer vision. In other words, it can separate different
internet to index and search for content by following links objects in a image or a video. You give it a image, it gives
and exploring.A web scraper is a specialized tool designed you the object bounding boxes, classes and masks.We use a
to accurately and quickly extract data from a web page. custom data set of guns to train this model .
Web scraper used here collects only images.First, a scraper
unique to your project, designed specifically to target and C. Masking
extract the data you want from the websites you want it A mask is applied to the object to produce the
from. censored image.
IJISRT20JUN709 www.ijisrt.com 891
Volume 5, Issue 6, June – 2020 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
This image is then given back to the browser to be A. Training Phase
viewed by the user. In this phase we create a dataset and annotate
them.The dataset contains almost 250 images. Validation
III. EXPERIMENTAL ANALYSIS set contains almost 100 images.
The image censoring section of the proposed system
consists of two parts testing phase and training phase.The
image is acquired through web browser.
Fig 1:- Dataset
B. Testing Phase
In this phase we evaluate the performance of the model. We adopt image-centric training. Images are resized such that their
scale (shorter edge) is 800 pixels. Each mini-batch has 2 images per GPU and each image has N sampled RoIs, with a ratio of 1:3
of positive to negatives.
Fig 2:- Test Phase Evaluation
IV. SOFTWARE IMPLEMENTATION D. Keras API
The Keras functional API used create models that is
A. Python more flexible than other API’s. This functional API used to
It used for processing and visualization of the data.It handle non-linear topology, shared layers, and multiple
is used for initial segmentation and generation of attribute inputs or outputs.
tables for data processing. It is also used to create
webscraper E. Anaconda
It is used to set up a platform for creating the object
B. Tensor flow detection model.It is integrates tensorflow and keras api
The fetched image is analyzed based on the factors into the training model.
such as slope,NDVI etc and these images are processed for
further scanning for predicting the output. V. HARDWARE IMPLEMENTATION
C. MS COCO Dataset 1 TB Hard disk
It is used for initial segmentation and detection.The 8 GB RAM minimum
pre trained model on MS COCO forms the basis for the Processor intel i5 or above
later model. GPU gtx 1050 or above
3 GB minimum vram
IJISRT20JUN709 www.ijisrt.com 892
Volume 5, Issue 6, June – 2020 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
VI. FUTURE SCOPE polishing and performance improvement can be done in
future to make it more appealing to general public.
The project in the future, can be further expanded by
improving the detection accuracyand reducing load VII. RESULT
times.Few ways it could be achieved is by Improving
model accuracy by adding more images to data set,reducing The model shows almost 80% accuracy in test
the dependency on on board hardware,Implementing the performed. We used cross validation for training
software entirely on a SaaS platform to make it more dataset.This software provides a more advanced to deal
flexible, adding more user friendly features.Much more with explicit images in webpages. The webpage with this
software running in background is shown below.
Fig 3:- Webpages after censoring.
VIII. CONCLUSION REFERENCES
The proposed system helps to censor violent and [1]. M. Andriluka, L. Pishchulin, P. Gehler, and B.
explicit images on webpages.This provides an efficient and Schiele. 2D human pose estimation: New benchmark
easy way to protect children from this type of content in and state of the art analysis. In CVPR, 2014.
internet.Also this project provides a platform for further [2]. P. Arbela ́ez, J. Pont-Tuset, J. T. Barron, F. Marques,
development of automatic censoring of unwanted content and J. Malik. Multiscale combinatorial grouping. In
on internet. CVPR, 2014.
[3]. A. Arnab and P. H. Torr. Pixelwise instance
segmentation with a dynamically instantiated network.
In CVPR,2017.
[4]. M. Bai and R. Urtasun. Deep watershed transform for
in- stance segmentation. In CVPR, 2017.
IJISRT20JUN709 www.ijisrt.com 893
Volume 5, Issue 6, June – 2020 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
[5]. S. Bell, C. L. Zitnick, K. Bala, and R. Girshick.
Inside- outside net: Detecting objects in context with
skip pooling and recurrent neural networks. In CVPR,
2016.
[6]. Z. Cao, T. Simon, S.-E. Wei, and Y. Sheikh. Realtime
multi- person 2d pose estimation using part affinity
fields. In CVPR, 2017.
[7]. M. Cordts, M. Omran, S. Ramos, T. Rehfeld, M.
Enzweiler, R. Benenson, U. Franke, S. Roth, and B.
Schiele. The Cityscapes dataset for semantic urban
scene understanding. In CVPR, 2016.
[8]. J. Dai, K. He, Y. Li, S. Ren, and J. Sun. Instance-
sensitive fully convolutional networks. In ECCV,
2016.
[9]. J. Dai, K. He, and J. Sun. Convolutional feature
masking for joint object and stuff segmentation. In
CVPR, 2015.
[10]. J. Dai, K. He, and J. Sun. Instance-aware semantic
segmen- tation via multi-task network cascades. In
CVPR, 2016.
