BACKGROUND REMOVAL OF VIDEO IN REALTIME
Arya Khanorkar
Yeshwantrao Chavan College of Engineering, Nagpur, Maharashtra, (India).
Bhavika Pawar
Yeshwantrao Chavan College of Engineering, Nagpur, Maharashtra, (India).
Diksha Singh
Yeshwantrao Chavan College of Engineering, Nagpur, Maharashtra, (India).
Kritika Dhanbhar
Yeshwantrao Chavan College of Engineering, Nagpur, Maharashtra, (India).
Nikhil Mangrulkar
Yeshwantrao Chavan College of Engineering, Nagpur, Maharashtra, (India).
E-mail: mangrulkar.nikhil@gmail.com
Reception: 17/11/2022 Acceptance: 02/12/2022 Publication: 29/12/2022
Suggested citation:
Khanorkar, A., Pawar, B., Singh, D., Dhanbhar, K., y Mangrulkar, N. (2022). Background removal of video in
realtime. 3C TIC. Cuadernos de desarrollo aplicados a las TIC, 11(2), 195-206. https://doi.org/
10.17993/3ctic.2022.112.195-206
https://doi.org/10.17993/3ctic.2022.112.195-206
3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529
Ed. 41 Vol. 11 N.º 2 August - December 2022
195
ABSTRACT
Background removal for video is a computer-vision based system to remove the background from a
video with ease. Creating a professional background when at home, i.e., not in a very professional
environment, can be a tedious task. Not everyone has time to learn editing and the technicalities
involved in having an entire setup for creating a sophisticated background and it is not practical that
normal people buy green screens or blue screens just for their everyday formal meets. Our goal is to
create a quick and easy solution to that by removing background in real time while also maintaining
the quality of the call, having the additional benefit of adding custom backgrounds and enabling users
to add effects like adjust lighting, contrast etc., we are combining all 4-5 steps in 1 single step.
SelfieSegmentation module of Mediapipe helps us achieve this. The Selfie Segmentation API creates
an output mask from an input image. The mask will be the same size as the input image by default. A
float integer with a range of [0.0, 1.0] is assigned to each pixel of the mask. The higher the confidence
that the pixel depicts a person, and vice versa, the closer the number is to 1.0.
KEYWORDS
Categories and Subject Descriptors: G.4 [Mathematics of Computing]: Mathematical Software - User
Interfaces; H5.2 [Information Interfaces and Presentation]: User Interfaces - User-centered design;
Interaction styles; Theory and methods.
Background Removal, Online Meetings, Professional Background, Streaming.
https://doi.org/10.17993/3ctic.2022.112.195-206
3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529
Ed. 41 Vol. 11 N.º 2 August - December 2022
196
1. INTRODUCTION
This pandemic has made it clear that virtual meets are here to stay. They are the new normal. But often
virtual meetups from home fail to create the professional atmosphere due to backgrounds in the video.
The professionalism required can be achieved by proper background in videos and minimal
disturbances. Unnecessary and visually unpleasing objects need to be removed. This can be achieved
by background removal, changing background color and adding virtual backgrounds. Background
removal can help avoid the cumbersome task of arranging a good backdrop in some cases even
eliminates the use of green screens which are used by many youtubers. It helps add to the overall
aesthetics and pleasantness of video conferences or even live streams.
The technique of mimicking human intelligence in robots programmed to think and behave like
humans is known as artificial intelligence (AI). Computer vision is a branch of artificial intelligence
(AI) that allows computers and systems to extract useful information from digital photos, videos, and
other visual inputs and act or make recommendations based on that data. A computer vision technique
called object detection is used to locate and identify objects in pictures and movies. Using this type of
identification and localization, object detection can be used to count the items in a scene, locate and
track them precisely, and accurately label them.
Image segmentation is the technique of dividing a digital image into numerous pieces for use in digital
image processing and computer vision (sets of pixels, sometimes known as image objects). Various
portions of a movie can be discovered using object detection and image segmentation, and different
adjustments can be performed to each of the components. We'll use these ideas to remove the
backdrop from a video using AI.
Open Broadcaster Software (OBS) helps you set your scene as a virtual camera to which we will be
feeding our video directly in real time using pyvirtualcam and mediapipe. We'll use Google Meet to
recognize OBS as a video source and output it as a virtual camera, resulting in enhanced quality of
video calls.
2. REVIEW OF LITERATURE
2.1 PRESENT SYSTEM
The apps that currently provide such a feature for removing the background of a video in real time
tends to worsen the quality of the video and do not provide us with smooth edges of the main object
which is not ideal for a professional use. Also, the process to remove or change the background in
these apps consists of various steps to be followed, sometimes we are even required to cut the video
call and rejoin with a changed background. Through our project we have tried to overcome these
problems by providing a system which removes the background while maintaining a proper quality of
the video with fps greater than 30. And we also provide an easy way to change the backgrounds by
just clicking a button on the keyboard.
Grzegorz Szwoch, in his paper presented, implementation of a background subtraction algorithm using
the OpenCL platform. The algorithm works using a live stream of video frames from an on-line
surveillance camera. A host machine and a parallel computing device are used to execute the
processing. The research focuses on optimizing OpenCL algorithm implementation for GPU devices
by taking into consideration specific GPU architecture aspects including memory access, data
transfers, and work group structure. The technique is designed to work on any OpenCL-enabled
device, including DSP and FPGA platforms. Several algorithm optimizations are presented and tested
on a variety of devices with variable processing power. The work's major goal is to figure out which
optimizations are required for on-line video processing in the surveillance system.
A relatively inexpensive background subtraction method is proposed by Hasup Lee et al., in their
study employing background sets with im-age- and color-space reduction. Background sets are used to
https://doi.org/10.17993/3ctic.2022.112.195-206
3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529
Ed. 41 Vol. 11 N.º 2 August - December 2022
197
recognize objects from dynamic backdrops like waves, trees, and fountains. The image space is
decreased to handle jittered and unstable frames, such as those from handheld mobile devices. The
color space is shrunk to account for color noise, such as the scattered RGB values from a digital
camera. To reduce expenses, a combination of color-space reduction and hash-table look-up operations
is used. The results, when compared to other methods, suggest that the proposed technology is
feasible: it may also be used in mobile or embedded environments.
S. Joudaki, et al., in their paper, they presented a comparison of numerous existing background
subtraction methods, ranging from basic background subtraction to more complicated providential
techniques. The purpose of this research is to provide an overview of the advantages and
disadvantages of commonly utilized approaches. The approaches are compared based on how much
memory they demand, how long they take to compute, and how well they handle different types of
films. Finally, other criteria such as processing time and memory needs were used to compare the
existing approaches. Baoxin Li, et al., in their paper they proposed a video background replacement
algorithm, this is based on adaptive background modelling and background subtraction. It can be
accomplished with a pre-recorded background scene image rather of a blue screen. Identifying
statistical outliers in respect to a specific background is the challenge. A two- pass approach is utilized
to modify initial segmentation based on statistics about a pixel's vicinity, which lowers false positives
in the background area while raising detection rates for foreground objects. Experiments with real
image sequences, as well as comparisons with other existing approaches, are shown to demonstrate the
benefits of the proposed methodology.
S. Brutzer, et al., in their paper, presented one of the most important approaches for automatic video
analysis, particularly in the field of video surveillance, is background subtraction. Despite their
usefulness, reviews of recent background removal algorithms in relation to video surveillance
challenges include several flaws. To address this problem, we must first identify the major obstacles to
background subtraction in video surveillance. We then evaluate the performance of nine back-ground
subtraction algorithms with post-processing depending on how well they over-come those challenges.
As a result, a fresh evaluation data set is presented that includes shadow masks and precise ground
truth annotations. This enables us to offer a thorough evaluation of the advantages and drawbacks of
various background sub-traction techniques.
In their study, R. J. Qian et al., presented an algorithm for altering video backgrounds without a blue
screen physically. Pre-recording a backdrop image of the scene free of any foreground objects is
required for the operation. Based on the color difference between the pixels in an input frame and their
corresponding pixels in the background image, the method computes a probability map that contains
the likelihood for each pixel to be classified into the foreground or background. The probability map is
further improved using anisotropic diffusion, which reduces classification mistakes without adding a
lot of artefacts. The foreground pixels from the input frames are then feathered onto a brand- new
background video or image based on the enhanced probability map to create the output video. The
method requires only a little amount of CPU resources and is designed to work in real time.
Experiment findings are also reported.
A. Ilyas, et al., in their paper, presented a Modified Codebook Model-Based Real Time Foreground-
Background Segmentation. The initial step in object tracking is the essential process of segmenting the
scene in real time into the foreground and background. beginning with the codebook approach.
Authors suggested certain changes that show notable improvements in the majority of the typical and
challenging conditions. For accessing, removing, matching, and adding codewords to the codebook as
well as moving cached codewords into the codebook, they included frequency options. They also
suggest an evaluation procedure based on receiver operating characteristic (ROC) analysis, precision
and recall methodology, to impartially compare various segmentation techniques. Authors suggested
expressing the quality factor of a method as a single value based on a harmonic mean between two
related features or a weighted Euclidean distance.
Rudolph C. Baron, et al., in their paper, presented a solution for managing a video conference. When
establishing a video conference with a second person, a first participant can choose from among
https://doi.org/10.17993/3ctic.2022.112.195-206
3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529
Ed. 41 Vol. 11 N.º 2 August - December 2022
198
several stored virtual backgrounds and use that background. One or more characteristics of the first
and/or second participant, one or more characteristics of the video conference, and/or similar
considerations may be used to choose the virtual background. The virtual backgrounds can be used,
for instance, to provide people outside of a company organization a desired perception, message, and/
or the like while they communicate with its employees via video conferencing. The virtual background
can incorporate static image data, live or recorded video feeds, static business entity web pages, and
dynamic business entity web pages.
Jian sun, et al., Effective techniques and approaches in a video sequence isolate the focus from the
background, according to their paper. In one instance, a system creates an accurate real-time backdrop
cut of live video by reducing the background contrast while maintaining the contrast of the
segmentation boundary itself. This method enhances the border between the foreground and
background images. The live video may then combine the fragmented foreground with another
background. An adaptive background color mixture model can be used by the system to distinguish
foreground from background more effectively when there are changes in the backdrop, such as camera
movement, lighting changes, and the movement of small objects in the background.
Juana E. Santoyo-Morales, et al., in their paper presented a Background sub-traction models based on
a Gaussian mixture have been widely employed in a range of computer vision applications for
detecting moving objects. Background sub-traction modelling, on the other hand, remains a challenge,
especially in video sequences with dramatic lighting changes and dynamic backdrops (complex
backgrounds). The goal of this research is to make background subtraction models more resilient to
complicated situations. The following enhancements were proposed as a result: Redefining the model
distribution parameters (distribution weight, mean, and variance) involved in the detection of moving
objects; enhancing pixel classification (background/foreground) and variable update mechanisms
using a new time-space dependent learning rate parameter; and c) substituting a new space-time
region-based model for the pixel-based model that is currently used in the literature.
According to Yiran Shen et al., background subtraction is a typical first step in many computer vision
applications, including object localization and tracking. Its objective is to pick out the moving parts of
a scene that match to the important things. Researchers in the field of computer vision have been
working to increase the reliability and accuracy of such segmentations, but most of their techniques
require a lot of computation, making them unsuitable for our target embedded camera platform, which
has a much lower energy and processing capacity. In order to create a new background subtraction
method that overcomes this issue while retaining an acceptable level of performance, authors added
Compressive Sensing (CS) to the often-used Mixture of Gaussian. The results imply that their
technique can significantly reduce the eventual time taking.
Semi-supervised video object segmentation should be considered, which is the process of creating
precise and consistent pixel masks for objects in a video sequence based on ground truth annotations
from the first frame, according to a suggestion made by Jonathan Luiten et al. To do this, they
provided the PReMVOS algorithm (Proposal- generation, Refinement and Merging for Video Object
Segmentation). The method separates the problem into two steps to specifically address the difficult
issues related to segmenting multiple objects across a video sequence: first, generating a set of precise
object segmentation mask proposals for each video frame; and second, choosing and merging these
proposals into precise and object tracks that are pixel-wise and consistently timed inside a video
sequence.
Thuc Trinh Le, et. al., demonstrated a method for removing items from videos. The technique simply
requires a few input strokes on the first frame that roughly delineate the deleted objects. Authors
claims that this is the first method which enables semi-automatic object removal from videos with
intricate backgrounds. The following are the main phases in their system: Segmentation masks are
improved after setup and then automatically distributed throughout the film. Video inpainting
techniques are then used to fill in the gaps. Authors claim that their system can handle several,
potentially intersecting objects, complex motions, and dynamic textures. As a result, a computational
https://doi.org/10.17993/3ctic.2022.112.195-206
3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529
Ed. 41 Vol. 11 N.º 2 August - December 2022
199
tool that can automate time-consuming manual tasks for editing high-quality videos has been
developed.
Thanarat H. Chalidabhongse, et. al., in their paper, showed how to use color pictures to detect moving
items in a static background scene with shading and shadows. They created a reliable and efficient
background subtraction method that can deal with both local and global lighting variations, such as
shadows and highlights. The approach is based on a proposed computational color model that
distinguishes between the brightness and chromaticity components. This technology has been used to
create real-world image sequences of both indoor and outdoor locations. The results, which
demonstrate the system's performance, are also provided, as well as several speed-up techniques used
in their implementation.
Yannick Benezeth, et. al., in their paper, presented comparison of different state-of-the- art background
subtraction approaches is presented. On several movies containing ground truth, there have been
developed and tested methods ranging from straightforward background subtraction with global
thresholding to more complex statistical algorithms. The purpose is to lay a solid analytic foundation
on which to highlight the benefits and drawbacks of the most extensively used motion detection
methods. The approaches are compared in terms of their ability to handle various types of videos,
memory requirements, and computing effort. A Markovian prior, along with several postprocessing
operators, are also considered. Most of the films are from modern benchmark collections and highlight
a range of issues, including low SNR, background motion in many dimensions, and camera jitter.
Yi Murphey, et. al., in their paper, describes their work on image content-based indexing and retrieval,
which is an important technique in digital image libraries. Image features used for indexing and
retrieval in most extant image content-based approaches are global, meaning they are computed over
the full image. Background features can readily be mistaken for object features, which is the
fundamental drawback of retrieval techniques based on global picture features. Users typically refer to
the color of a particular object or objects of interest in an image while searching for photos using color
attributes. The technique described in this article uses color clusters to analyze image backgrounds.
After being identified, the background regions are deleted from the image indexing process, so they
won't interfere with it anymore. Three main calculation processes make up the algorithm: fuzzy
clustering, color picture segmentation, and background.
Zhao Fei Li, et. al., in their paper, presented background noise removal is a key stage in the picture
processing and analysis process. Researchers use a variety of techniques to remove background noise
from images. For instance, grey threshold techniques are frequently used to eliminate noises that have
a strong contrast to the object of interest. However, there are a lot of noises in the grey scale that don't
change as the interesting objects do. These noises cannot be reduced using the grey level-based noise
removal technique, but the contour feature is excellent at doing so. The contour feature-based image
background removal approach depends on the contour model. The contour characteristic of the interest
items is modelled using a revolutionary method proposed in this study. A unique background noise
with the same grey level as the background noise is completely eradicated using this method.
Thuc Trinh Le, et. al., demonstrated a method for eliminating objects from videos in their study. A few
strokes in at least one frame are all that are needed for the technique to roughly delimit the items to be
eliminated. These undeveloped masks are then polished and automatically broadcast throughout the
video. The corresponding areas are synthesized again using video inpainting methods. Authors claim
that their system is capable of navigating several, perhaps crossing objects, intricate motions, and
dynamic textures. As a result, a computational tool has been created for editing high-quality videos
that can take the place of laborious human work.
3. PROPOSED TECHNIQUE
The main objective of the proposed technique is to create a simpler process of removing background
from a live or saved video, to simplify the process of creating an aesthetic background. Our project
https://doi.org/10.17993/3ctic.2022.112.195-206
3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529
Ed. 41 Vol. 11 N.º 2 August - December 2022
200
allows the user to completely remove the background or put a different background and be able to
switch between multiple backgrounds or colors.
The main objective of our project is to eliminate the need to physically change the arrangement of
room for a better and professional background. Thus, providing users an easy way of applying or
changing backgrounds while they are in some online meet or live streams while maintaining the
quality of the video.
3.1 Flow of Proposed Technique
Fig. 1. Flowchart of the proposed system.
Fig. 1 shows the flow of the proposed system, starting with the accessing of the live video from
webcam to the removal/changing of background and feeding the output to applications like
google meet using OBS Virtual Camera.
3.2. IMPLEMENTATION
Our goal was to remove the background in real-time and with FPS more than 30.
3.2.1. STARTING THE WEBCAM
We should be able to access the webcam by simply running the code so that the video i.e., real
time video can be directly taken as input. A computer vision library is called OpenCV (Open-
Source Computer Vision). with a variety of image and video manipulation tools. The OpenCV
library can be used to manipulate films in a variety of ways. To capture a video, you'll need a
VideoCapture object. The index of the device or the video file’s name is stored in VideoCapture.
The device index is just a number that identifies which of the camera device is being used.
Syntax: capt = cv2.VideoCapture(0)
Now a pop-up window will open if we have a webcam. We have set the frame size to 640 X 480.
Therefore, background-replacing images should be 640 x 480, which is the same size as the
frame.
Creating a dataset for background images Make a folder called 'BackgroundImages' inside the
project directory. You can download and store any image, or any number of images, in this
directory.
https://doi.org/10.17993/3ctic.2022.112.195-206
3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529
Ed. 41 Vol. 11 N.º 2 August - December 2022
201
3.2.2. BACKGROUND REMOVAL
We have used the SelfieSegmentationModule from cvzone package which uses OpenCV and
Mediapipe libraries at its core and makes AI operations on videos and images very easy.
SelfieSegmentation is a technique for removing the frame's background and replacing it with
photos from our directory. It is based on MobileNetV3 but has been tweaked to be more efficient.
It uses a 256x256x3 (HWC) tensor as input and outputs a 256x256x1 tensor as the segmentation
mask. Before feeding it into the ML models, MediaPipe SelfieSegmentation automatically resizes
the input image to the necessary tensor dimension. We use the webcam for input and frame
width should be set to 640 x
480. Then we utilize the cvzone to execute SelfiSegmentation(), which carries out object
identification, image segmentation, and ultimately background removal. The output frames can
show the frames per second (fps) using the FPS() method.
Syntax: seg = SelfiSegmentation() Setfps = cvzone.FPS()
SelfieSegmentation() converts the image into RGB and sends it to the Selfie Segmentation model
to process and then it checks if the image is colored; if yes it changes the color of the background
and if not, it then changes the color of the image. As a result, we can see the background
successfully removed.
3.2.3. STORE BACKGROUND IMAGES IN A LIST
Then, after creating a list of every image in the BackgroundImages folder, we iterate through it,
reading each one and adding it to an empty list. At the beginning, the index is set at zero.
3.2.4. REPLACE BACKGROUND WITH DESIRED BACKGROUND
The frames are read from the camera using a while loop, and the background is then removed
from the frames using the seg.removeBG() method and replaced with images from the directory.
The camera's image frame (img), the directory's collection of photos, and with an index of image
(imgList[indexImg]), are all passed to the seg.removeBG() function along with the threshold. To
improve the edges, we additionally modify the threshold setting.
3.2.5. FUNCTIONALITY TO CHANGE BACKGROUND USING KEYBOARD SHORTCUTS
Using cvzone.stackImages, we stack the images and retrieve the output of the frames or
background-replaced image. Then, by means of a straightforward if statement, we assign keys to
change the background. The principle is to sequentially remove the indexes according to the key
that was pressed to display the image for the resulting index. This lets you change the
backgrounds quickly.
3.2.6. SEND THE FRAMES TO OBS VIRTUAL CAMERA
We then send the resulting frames to OBS Virtual Camera using Pyvirtualcam(). It sends frames
to a virtual camera from Python. The OBS virtual camera is detected by various platforms, and
we have our very own background removal with whichever backgrounds, or no backgrounds, as
required.
4. RESULTS AND DISCUSSIONS
With our proposed implementation, we successfully removed the background and add any
other desired background in real time with FPS more than 30.
https://doi.org/10.17993/3ctic.2022.112.195-206
3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529
Ed. 41 Vol. 11 N.º 2 August - December 2022
202
Fig. 2. Removal of background in real time.
Fig. 2. shows that the left half of the image has normal real-time video with background and the
right side has video with no background with FPS 34.
Fig. 3. Background changed in real time successfully with FPS = 50.
Fig. 3. shows that left side of the image has normal real-time video, and the right side has video
with desired background with FPS 50.
Fig. 4. Background changed of pre-recorded video successfully with FPS= 61.
Fig. 4. shows that left side of the image has normal pre-recorded video, and the right side has the
video with desired background with FPS 61.
https://doi.org/10.17993/3ctic.2022.112.195-206
3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529
Ed. 41 Vol. 11 N.º 2 August - December 2022
203
Fig. 5. Background changed in real time successfully on Google Meet.
Fig. 5. shows that we were able to change the background of our live video in a Google
meet.
Fig. 6. Background changed in real time successfully and able to configure in Google Meet.
Fig. 6. shows that we were able to change the background of our live video as well as configure
the features such as brightness, contrast, saturation, etc. of our real time video in a Google meet.
4. CONCLUSION
We have implemented a computer-vision based system to remove the background from a video in
real-time which enabled creating a professional background when not in a very professional
environment. Our goal was to create solution for removing background in real time while also
maintaining the quality of the call, having the additional benefit of adding custom backgrounds
and enabling users to add effects like adjust lighting, contrast etc. We used SelfieSegmentation
module of Mediapipe in this implementation. Our results shows that our technique successfully
removed backgrounds from live videos as well as prerecorded videos at frame rate between 30
and 60. We also changed background of videos in reals time and prerecorded videos seamlessly.
Our implementation also worked very well on streaming platforms like Google Meet.
REFERENCES
[1] A. ILYAS, M. SCUTURICI AND S. MIGUET, 2009, "Real Time Foreground-Background
Segmentation Us - ing a Modified Codebook Model," 2009 Sixth IEEE International
https://doi.org/10.17993/3ctic.2022.112.195-206
3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529
Ed. 41 Vol. 11 N.º 2 August - December 2022
204
Conference on Advanced Vi -deo and Signal Based Surveillance, pp. 454-459, doi:
10.1109/AVSS.2009.85.
[2] BAOXIN LI AND M. I. SEZAN, 2001, "Adaptive video background replacement," IEEE
International Conference on Multimedia and Expo, 2001. ICME 2001., pp. 269-272, doi:
10.1109/IC ME.2001.1237708.
[3] G. SZWOCH, 2014, "Parallel background subtraction in video streams using OpenCL on
GPU platforms," 2014 Signal Processing: Algorithms, Architectures, Arrangements, and
Applications ( SPA), pp. 54-59.
[4] H. LEE, H. KIM AND J. KIM, 2016, "Background Subtraction Using Background Sets With
Image- and Color- Space Reduction," in IEEE Transactions on Multimedia, vol. 18, no. 10,
pp.2093- 2 103 , doi: 10.1109/TMM.2016.2595262.
[5] JIAN SUN HEUNG-YEUNG, SHUM XIAOOU AND TANG WEIWEI ZHANG, 2009,
“Background Removal In A Live Video,” Assigned to Microsoft Technology Licensing, LLC.
[6] JONATHON LUITEN, PAUL VOIGTLAENDER AND BASTIAN LEIBE, 2019, "PReMVOS:
Proposal-
[7] Generation, Refinement and Merging for Video Object Segmentation” In book: Computer
Vision – ACCV 2018 (pp.565-580), DOI:10.1007/978-3-030-20870-7_35
[8] JUANA E. SANTOYO-MORALES AND ROGELIO HASIMOTO BELTRAN, 2014, “Video
background
[9] Subtraction in Complex Environments” in Journal of Applied Research and Technology
12(3):527-427, DOI:10.1016/S1665-6423(14)71632-3
[10] R. J. QIAN AND M. I. SEZAN, 1999, "Video Background Replacement without a Blue
Screen". Proc. of IEEE International Conference on Image Processing. Kobe, Japan.
[11] RUDOLPH C. BARON, ANDREW R. JONES, MICHEAL M. MASSIMI, KEVIN C.
MCCONNELL, 2014, “Background Replacement for Video Conferencing”, Application filed
by International Business Machines Corp.
[12] S. BRUTZER, B. HÖFERLIN AND G. HEIDEMANN, 2011, "Evaluation of background
subtraction techniques for video surveillance," CVPR 2011, pp. 1937-1944, doi: 10.1109/
CVPR.2011.59955 08.
[13] S. JOUDAKI, M. S. BIN SUNAR AND H. KOLIVAND, 2015, "Background subtraction
methods in video streams: A review," 2015 4th International Conference on Interactive
Digital Media (ICIDM), pp. 1-6, doi: 10.1109/IDM.2015.7516329.
[14] CHALIDABHONGSE, THANARAT & HARWOOD, DAVID & DAVIS, LARRY, 1999,
“A
statistical approach for real-time robust background subtraction and shadow detection”, IEEE
ICCV.
[15] THUC TRINH LE, ANDRÉS ALMANSA, YANN GOUSSEAU AND SIMON MASNOU,
2019, "Object removal from complex videos using a few annotations", in Computational
Vidual Media 5(3) August 2019, DOI:10.1007/s41095-019-0145-0, Project: Video Inpainting
[16] THUC TRINH LE, ANDRÉS ALMANSA, YANN GOUSSEAU AND SIMON MASNOU,
2018, “Removing objects from videos with a few strokes" in Conference: SIGGRAPH Asia
2018 Technical Briefs, DOI:10.1145/3283254.3283276.
[17] YANNICK BENEZETH, PIERRE-MARC JODOIN, BRUNO EMILE, HÉLÈNE
LAURENT
AND CHRISTOPHE ROSENBERGER, 2010, " Comparative study of
background subtraction algorithms" in Journal of Electronic Imaging
19(3):033003-033003,
DOI:10.1117/1.3456695
https://doi.org/10.17993/3ctic.2022.112.195-206
3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529
Ed. 41 Vol. 11 N.º 2 August - December 2022
205
[18] YI LU AND HONG GUO, 1999, "Background removal in image indexing and retrieval,"
Proceedings 10th International Conference on Image Analysis and Processing, pp. 933-938, doi:
10.1109/ICIAP.1999.797715.
[19] YIRAN SHEN,WEN HU, MINGRUI YANG AND JUNBIN LIU, 2012, "Efficient background
subtraction for tracking in embedded camera networks" DOI:10.1145/2185677.2185698
[20] ZHAO FEI LIV AND JIANG QING WANG, 2014, “Image Background Removal by Contour
Feature" in Advanced Materials Research 926-930:3050-3053, DOI:10.4028/
www.scientific.net/ AMR.926-930.3050
https://doi.org/10.17993/3ctic.2022.112.195-206
3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529
Ed. 41 Vol. 11 N.º 2 August - December 2022
206
