Computer Science and Information Technologies

Vol. 5, No. 2, July 2024, pp. 130~139

ISSN: 2722-3221, DOI: 10.11591/csit.v5i2.pp130-139  130

Video shot boundary detection based on frames objects

comparison and scale-invariant feature transform technique

Noor Khalid Ibrahim, Zinah Sadeq Abduljabbar

Department of Computer Science, College of Science, Mustansiriyah University, Baghdad, Iraq

Article Info ABSTRACT

Article history: The most popular source of data on the Internet is video which has a lot of
information. Automating the administration, indexing, and retrieval of movies
Received Dec 12, 2023 is the goal of video structure analysis, which uses content-based video
Revised Feb 24, 2024 indexing and retrieval. Video analysis requires the ability to recognize shot
Accepted Mar 4, 2024 changes since video shot boundary recognition is a preliminary stage in the
indexing, browsing, and retrieval of video material. A method for shot
boundary detection (SBD) is suggested in this situation. This work proposes
Keywords: a shot boundary detection system with three stages. In the first stage, multiple
images are read in temporal sequence and transformed into grayscale images.
Frames correlation Based on correlation value comparison, the number of redundant frames in
Object comparison the same shots is decreased, from this point on, the amount of time and
Shot boundary computational complexity is reduced. Then, in the second stage, a candidate
Video analysis transition is identified by comparing the objects of successive frames and
Video segmentation analyzing the differences between the objects using the standard deviation
metric. In the last stage, the cut transition is decided upon by matching key
points using a scale-invariant feature transform (SIFT). The proposed system
achieved an accuracy of 0.97 according to the F-score while minimizing time
consumption.

This is an open access article under the CC BY-SA license.

Corresponding Author:
Noor Khalid Ibrahim
Department of Computer Science, College of Science, Mustansiriyah University
Baghdad, Iraq
Email: noor.kh20@ uomustansiriyah.edu.iq

1. INTRODUCTION
The vast amount of video content on the internet makes it challenging to develop effective indexing
and search strategies for managing video data. Content-based video retrieval is emerging as a trend in video
retrieval systems, while conventional methods like video compression and summarizing aim for minimal
storage requirements and maximum visual and semantic accuracy [1]. Given that video is the most
sophisticated sort of multimedia data, it includes information about the target's mobility within the scene as
well as information about the objective world changing with time [2].
Two modules can be approximately regarded in video segmentation which are video object
(foreground/background) segmentation, and video semantic segmentation [3]. Video segmentation, also known
as shot boundary detection (SBD), involves breaking the video up into meaningful scenes so that the essential
feature(s) may be found in each scene through analysis [4]. A cut is a sudden change in the shot that takes place
inside a single frame. A fade is a gradual alteration in brightness that often begins or ends with a completely
dark frame. Frames inside the transition show one image overlaid on the other during a dissolve, which happens
as the images of the first shot go darker and the images of the second shot get brighter [1]. The primary

Journal homepage: http://iaesprime.com/index.php/csit

Comput Sci Inf Technol ISSN: 2722-3221  131

difficulties in shot boundary recognition are movements of the camera and objects since these can significantly
change the video content, producing an effect akin to transition effects and leading to inaccurate shot transition
detection [5].
Numerous studies have addressed video segmentation, Hong Shao et al. [6] utilized a combination of
a color histogram with Hue Saturation Value (HSV) and features of histogram of gradient (HOG) to effectively
detect abrupt shot changes in videos. In [3] This work proposes a shot boundary detection approach based on
the scale-invariant feature transform (SIFT). Using a top-down search strategy, the initial phase of this
approach compares the ratio of matched features derived by SIFT for each RGB channel of video frames to
locate transitions. The boundaries' locations are shown in the overview stage. Second, to ascertain the kind of
transition, a moving average computation is made.
In [7] The research aimed to use a multi-modal visual features-based SBD framework; the behaviors
of the visual representation are analyzed concerning the discontinuity signal. This used a candidate segment
selection strategy that does not compute the threshold; instead, it utilizes the discontinuity signal's cumulative
moving average to determine the shot boundary locations while disregarding the non-boundary video frames.
To differentiate between a candidate segment that is a cut transition and one that is a gradual transition,
including fade in/out and logo occurrence, the transition detection is carried out structurally.
In [8] the proposed temporal video segment representation formalizes video scenes as temporal
motion change data, determining motion modifications and cuts between scenes through optical flow character
changes. This reduces the issue to an optical flow-based cut detection problem, enhancing a pixel-based
representation. The proposed video segment representation divides temporal video segment points into cuts
and non-cuts.
In [9] the bag of visual word (BoVW) model, which splits the video into shots and keyframes, is the
basis for the segmentation model for videos that the study suggested. The BoVW model is employed in two
variants: the traditional BoVW and an expansion known as the vector of linearly aggregated descriptors
(VLAD). Keyframe feature vectors inside a sliding window of length L are used to calculate similarity. In [10]
The study presents a method for feature fusion and clustering technique (FFCT)-based video shot boundary
detection, which involves converting interval frames into grayscale images, extracting fingerprint and speed-
up robust features, fusion, and clustering them using a K-means algorithm. Linear discriminant analysis (LDA)
is introduced for cluster mapping, and features are chosen using density computation based on frame
correlation.
In [2] a novel algorithm for camera detection based on SIFT features was introduced in this study.
The proposed method involves the analysis of multiple frames of images in a sequential manner. Initially, the
images are converted into grayscale and divided into blocks. Subsequently, the dynamic texture of the film is
computed, and the correlation between the dynamic texture of adjacent frames and the matching degree of
SIFT features is determined. Based on these matching results, pre-detection outcomes are obtained.
Idan et al. [11] proposed a fast video processing method for SBD. To reduce computing costs and
disturbances, the proposed SBD framework makes use of candidate segment selection with frame active area
and separable moments. Inequality criteria and adaptive threshold are used to exclude non-transition frames
and maintain candidate segments. Cut transition detection is done using machine learning statistics.
In [12] a practical SBD method was presented in the study, which uses average edge information for
gradual transition detection and gradient and color information for abrupt transition detection. Processing only
transition regions yield an average edge frame and reduces computational complexity. In [5] The proposed
method comprises two distinct stages. In the initial stage, projection features were employed to differentiate
between non-boundary transitions and candidate transitions that potentially encompass abrupt boundaries.
Consequently, only the candidate transitions were retained for further analysis in the subsequent stage. This
approach effectively enhances the speed of shot detection by minimizing the detection scope. In [13] An
effective SBD approach with several invariant properties was presented in this work. With the right mix of
invariant features, such as edge change ratio (ECR), color layout descriptor (CLD), and scale-invariant feature
transform (SIFT) key point descriptors, the accuracy level of SBD was increased.
According to the literature, many applications have been created to address the issue of shot boundary
detection in videos. These applications are performed based on various techniques to process the challenges in
SBD. This proposed SBD system has been achieved in three stages to improve its performance and try to
reduce the problem of object and camera motion, wherein the first stage the redundancy frames in the same
shots are reduced based on correlation value comparison, this stage yields minimizing time-consuming and
computation complexity. Then in the second stage candidate transition is determined by comparing the objects
of sequential frames, final stage the decision of the cut transition is made based on key points matching of SIFT
method. This proposed method aims to find the boundary frame of a shot with a cut transition between
consecutive shots accurately. The rest of the paper is organized as follows, section 2 explains the proposed
method, the experimental result, and the analysis demonstrated in section 3, followed by a conclusion in
section 4.
Video shot boundary detection based on frames objects comparison and scale… (Noor Khalid Ibrahim)
132  ISSN: 2722-3221

2. SBD PROPOSED METHOD

This proposed SBD system has been achieved in three stages, in the first stage, multiple images are
read in temporal sequence and transformed into grayscale images. Based on correlation value comparison, the
number of redundant frames in the same shots is decreased, and then, in the second stage, a candidate transition
is identified by comparing the objects of successive frames using the proposed method to extract frame image
objects. In the last stage, the cut transition is decided upon by matching key points using the SIFT approach.
The details of these stages are explained as follows:

2.1. Reduces redundancy stage

The multiple frames of input video are extracted as the first step, then converted into grayscale and
resized into 256*256. Some pre-processed operations are achieved on these frames to improve their quality
when the noise is removed by the wiener filter [14], and contrast is enhanced by histogram equalization [15].
The resulting frames are normalized in the range [0-1].
In one shot the consecutive frames have a very high similarity, and achieving the SBD process on
each pair of frames will be very time-consuming and computationally complex. So, to minimize this time and
complexity the redundancy frames in one shot have been reduced based on the measure of their correlation
value. The correlation value (r) of frames Fr(i) and Fr(i+1) and based on the threshold value (Th) identified
experimentally the frame Fr(i) is passed to the next stage, otherwise, frame Fr(i) is discarded as demonstrated
in (1). Where the correlation value is calculated as explained in (2) [16].

Passed to next stage r < Th

Fr(i)
discard otherwise (1)

∑𝑖(𝑥𝑖 −𝑥𝑚 )(𝑦𝑖 −𝑦𝑚)

𝑟= (2)
√(∑𝑖(𝑥𝑖 −𝑥𝑚 )2 √(∑𝑖(𝑦𝑖 −𝑦𝑚 )2

where 𝑥𝑖 denotes the pixel intensity in order ith of the first image, and 𝑦𝑖 demarcated the ith pixel
intensity of the second image, additionally, 𝑥𝑚 and 𝑦𝑚 is the mean intensity of first and second images
sequentially.

2.2. Selection of candidate transition stage

Candidate transition selection is performed based on comparison made on consecutive frame objects,
that means on frame image content. This image content extraction is achieved based on the proposed extraction
method as explained in Figure 1in this stage. As seen in the figure, the objects of the frame have been extracted
in two steps, which are the generation of the feature template, and extract the object, these steps are detailed as
following:

Figure 1. Frame objects extraction flowchart

2.2.1. First step (generate features template)

For each consecutive frame passed to this stage the template of features is generated when multiple
features are extracted and combined from each frame image. The selection of these multiple features must be
able to extract the objects of a frame image accurately, so in this proposed extraction method of this proposed
SBD algorithm, the multiple features are represented by the texture characteristics that yield information about
the local variability of the pixel's intensity values are recovered using the standard deviation filter (SD) [17] of
the 3-by-3 neighborhood around the consistent pixel. The value luminance grayscale of these processed frames

Comput Sci Inf Technol, Vol. 5, No. 2, July 2024: 130-139

Comput Sci Inf Technol ISSN: 2722-3221  133

is represented by channel L* in the L* a* b* color space [18] used as second feature. The L*a*b* typically
appears to be able to depict the colors to human vision. Additionally, because the RGB representation includes
a transition color between blue and green, the L*a*b* color representation compensates for the diversity in the
color distribution in the RGB color model [19]. For this reason, L*a*b* is taken into account along with its L*
value. These two feature matrices are then merged with the edge of the detected frame by a canny operator
which has the ability to recognize object boundaries in an image and object appreciation to create a feature
template. The following is how SD is calculated [20].
𝑁
1
𝜇𝑗 = ∑ 𝑥𝑗𝑖 (3)
𝑁
𝑖=1

𝑁
1
𝜎𝑗 = √ ∑(𝑥𝑗𝑖 − 𝜇𝑗 )2 (4)
𝑁
𝑖=1

2.2.2. Second step (objects extracted)

Utilize the k-means [21] algorithm with this created template to extract the objects from these
successive frames. A k-number group of data is gathered in order to use K-means. kmeans method consists of
two stages. In the first, the centroid is initialized, and in the second, the distance to the closest centroid is used
to identify which cluster the data point belongs to. Because of its ease of use and quick calculation, the k-means
clustering approach is widely utilized in clustering processes [22], which is the reason that it was chosen for
this phase. Consequently, the frame image object has been identified based on this proposed extraction method
with generated features template and K-means technique.
The frames' similarity has been measured based on the objects' comparison by dividing images of
objects of related sequential frames into 8×8 blocks, and then the entropy value of each block is calculated, in
turn, these entropies values are arranged into vectors of the length 64, which represent similarity measurement
vectors as explained in Figure 2, and then the standard deviation is calculated to differences between these two
entropies vectors of object images of consecutive frames when the value of standard deviation is nearest to
zero normal transition has been distinguished. According to the threshold (Thr) value perceived experimentally,
the abrupt transition has been a candidate, otherwise, normal transition has been detected as demarcated in (5).
Entropy value is determined as in (6) [23]. In turn, these candidate frames are passed to the third stage to make
a decision of abrupt transition.

Figure 2. Construct similarity measurement vector of object image

Abrupt transition candidate sd > Thr

Fri
Normal transition otherwise (5)

Let Fri represent the video frame with index i

Video shot boundary detection based on frames objects comparison and scale… (Noor Khalid Ibrahim)
134  ISSN: 2722-3221

𝐻𝑟 = − ∑ −𝑘 −𝑘
𝑔𝑟 log 2(𝑔𝑟 ) (6)

Where 𝑔𝑟−𝑘 denote distribution of assumed color space.

2.3. Transition decision stage

Making the right choice when deciding how to divide a video sequence into shots is mostly dependent
on selecting the right method. David adopted a scale-invariant feature transform SIFT [24]. The SIFT feature
has been used in this stage to determine the frame transition and its boundary because, given an image as input,
the SIFT descriptor generates a wide range of local feature vectors that are independent of image scaling and
rotation. SIFT is capable of precisely correlating two images [13]. In situations of abrupt transitions, when the
matching degree of the SIFT feature between the frames is low, neighboring frames are recognized as
belonging to different shots, which can better discern the moving objects in successive frames.

3. EXPERIMENTAL RESULTS AND ANALYSIS

Eight distinct videos from the standard dataset, TRECVid 2001 test data made existing on the open
video project and accessible at https://open-video.org, are used to assess the suggested method in this research.
These videos are referred to as Vid1 through Vid8. A comprehensive description of those input videos is
provided in Table 1. The ground truth value is determined by observing abrupt changes as seen by people. The
chosen videos contain a variety of aberrations, including lighting variations, viewpoint shifts, scaling, zooming,
rotation, and more.

Table 1. Description of input videos

Input Video name Time Duration Frames Abrupt
video In sec. number transition
Vid1 Free-for-all race at Charter Oak Park (Historical) 26 853 3
Vid2 New Indians, Segment 101 (Documentary) 131 3953 14
Vid3 New Indians, Segment 01 (Documentary) 56 1687 15
Vid4 Winning Aerospace, Segment 02 (Documentary) 65 1970 11
Vid5 Hidden Fury, segment 10 (Documentary) 33 1002 1
Vid6 Hurricanes, Segment 05 (Documentary) 115 3448 32
Vid7 The Miracle of Water, segment 05 (Documentary) 83 2314 1
Vid8 Winning Aerospace, Segment 04 (Documentary) 110 3318 18

3.1. Reduces redundancy stage

When the multiple frames of input video have been extracted, the frames images in the same shot
have a high similarity degree and when performing features extraction to extract objects from each frame image
results in time-consuming and computing complexity, so reducing the redundancy frames stage results in time-
consuming minimization as seen in Table 2 and Figure 3, for instance, the execution time was equivalent to
(111.4 seconds) when the second stage was applied to all of the vid1's frames, that means without similarity
frames reduction. as opposed to the execution time (44.41 seconds) when vid1 advanced to the lower
redundancy level, and so on to others videos as explained in this table that shows how much time each utilized
video takes.

3.2. Selection of candidate transition stage

Based on the motion of the object and/or the camera, shots may be categorized into four types: static
objects with static cameras, dynamic objects with static cameras, dynamic objects with dynamic cameras, and
dynamic objects with dynamic cameras [25]. Candidate transition selection is performed based on comparison
made on consecutive frames objects. This stage is achieved by comparison made to the extracted objects of
frames images based on the created features template by combining multiple features texture, edges, and L*
value of L*a*b* color space applied to the k-means technique.
The stage starts choosing potentially cut transition frames by examining the standard deviation to the
differences of vectors created from frames object blocks for similarity comparison after the frames Fri and
Fri+1 pass the first stage based on their correlation value measure. Table 3 and Figure 4 describe how the block
size of the frame image object is determined empirically, where Figure 4(a) explains the block size effect on
execution time and Figure 4(b) demonstrates the effect of block size on F-score. Vid3, Vid4, and Vid8 are
taken as examples to demonstrate that block size affects execution time and accuracy in this table. For this
investigation, 8*8 blocks with a 32*32 block size are more appropriate in this study.

Comput Sci Inf Technol, Vol. 5, No. 2, July 2024: 130-139

Comput Sci Inf Technol ISSN: 2722-3221  135

Table 2. Time consuming comparison

Videos Execution time Execution time
in Sec. (with reduction) in Sec. (without reduction)
Vid1 44.41 111.42
Vid2 224.94 785.83
Vid3 99.75 314.75
Vid4 129.07 363.44
Vid5 70.178 177.52
Vid6 320.34 566.40
Vid7 111.47 421.15
Vid8 201.23 679.02

Figure 3. Comparison in execution time

(a) (b)

Figure 4. Block size effect, (a) on execution time and (b) on F-score

Table 3. The effect of block size within 256*256 frame size

videos 4*4 blocks (64*64 block size) 8*8 blocks (32*32 block size)
Execution time F-score Execution time F-score
In Sec. In Sec.
Vid3 183.58 0.95 99.75 0.96
Vid4 148.92 0.90 129.07 1.00
Vid8 311.36 0.90 201.23 0.97

To explain the frame’s object extraction, for example with samples of frames that explained in
Figure 5, the frame objects extraction method steps demonstrate in Figure 6. The recovered combined features
(Texture, frame edge, and L* value of L*a*b* color space) from frames i and i+1 create the template features
for each one. The frame objects are then extracted for the frame similarity comparison using the k-means
approach. If identical objects are found in two consecutive frames, they are likely associated with the same
shot; if not, a cut shot transition is a possibility. The significant problem of object and camera movements can
be addressed by similarity discovery based on object comparison because the frame object is recognized where
it should be in the image of succeeding frames.

Video shot boundary detection based on frames objects comparison and scale… (Noor Khalid Ibrahim)
136  ISSN: 2722-3221

Figure 5. transitions examples

Figure 6. Example of consecutive frames objects extracted

This proposed object extraction method has been assessed for adopting in this proposed SBD
algorithim. According to Table 4 and Figure 7, which describe the information content as determined by the
entropy value that means the accuracy of extracting objects by the proposed extraction method of frame, in this
table some frames that apply extraction its objects from some different used videos are selected as samples for
evaluation. As a result of this evaluation explained in this table, and from the analysis of this evaluation, this
proposed object extraction operation has been adopted in this stage of the proposed SBD algorithm.

Figure 7. Object extraction accuracy using entropy

Table 4. Object extraction evaluation using entropy measure (Ent)

Vid2 Vid3 Vid5 Vid6 Vid7 Vid8
Fr. 397 398 262 263 432 432 82 83 568 569 1375 1376 517
Ent 0.926 0.634 0.890 0.985 0.660 0.969 0.998 0.989 0.968 0.992 0.979 0.956 0.884

Comput Sci Inf Technol, Vol. 5, No. 2, July 2024: 130-139

Comput Sci Inf Technol ISSN: 2722-3221  137

3.3. Transition decision stage

The SIFT properties are adopted in this stage for shot transition decision-making because when it
comes to rotation, and zoom, SIFT characteristics remain unaffected and it able to reflect the local variation of
moving object efficiently, and may be used to impartially characterize the image [2]. SIFT key points are
detected, features are extracted from candidate frames of video results from the previous stage, then feature
matching is performed. In features matching two features’ matrices of frame i, frame i+1 have been matched
using distance calculation results in a p-by-1 vector, where p represents the key point number that is detected.
And from the matched features shot transition decision-making, when the matching degree of the SIFT feature
between the frames is low, neighboring frames are recognized as belonging to different shots. Figure 8(a)
demonstrates features key point matching for frames in same shot, and Figure 8(b) from different shot.

(a) (b)

Figure 8. Frames shots feature key points matching, (a) frames in the same shot and (b) frames in a
different shot

As seen in the figure, due to comparable visual features, the similarity matching between two frames
in the same shot is typically high. Frames from diverse shots, however, lack visual uniformity. They therefore
have either little or no similarity matching.
Recall and precision are the key performance metrics of the suggested system that are typically
employed in the SBD process. The F1 score, which is the harmonic mean of precision and recall, is used in
this paper's evaluation along with these metrics [2]. The following formula can be used to compute these
metrics [5]:

𝑡𝑟𝑢𝑒
𝑅= (7)
𝑡𝑟𝑢𝑒 + 𝑚𝑖𝑠𝑠
𝑡𝑟𝑢𝑒
𝑃= (8)
𝑡𝑟𝑢𝑒 + 𝑓𝑎𝑙𝑠𝑒

2∗𝑃∗𝑅
𝐹 − 𝑠𝑐𝑜𝑟𝑒 = (9)
𝑃+𝑅

where True denotes accurate transition detection, False denotes inaccurate transition detection, and Miss
denotes missed transition detection. Table 5 demonstrates the accuracy with these metrics of this proposed
SBD algorithm.

Table 5. Efficiency of the proposed method

Video Recall Precision F-score
Vid1 1.00 1.00 1.00
Vid2 1.00 1.00 1.00
Vid3 0.93 1.00 0.96
Vid4 1.00 1.00 1.00
Vid5 1.00 1.00 1.00
Vid6 0.87 0.96 0.91
Vid7 1.00 1.00 1.00
Vid8 0.94 0.94 0.94
Average 0.96 0.98 0.97

Video shot boundary detection based on frames objects comparison and scale… (Noor Khalid Ibrahim)
138  ISSN: 2722-3221

3. CONCLUSION
By comparing frame image objects and using a scale-invariant feature transform SIFT feature with
the discard to the redundant frames of the same shot, the suggested SBD approach has been realized. Three
stages are involved in implementing this proposed system: first, the redundancy frames are reduced based on
their correlation value; this reduces computation complexity and time consumption; second, the candidate shot
transition and boundary are identified based on object comparison using proposed extraction method; this stage
can identify objects that where should be in the image of subsequent frames. The last step then uses the SIFT
feature to choose which of these candidate frames to select. The research demonstrates that this approach
minimizes false positives by utilizing SIFT matching key points, which are independent of the scale and
rotation of the image. Our method yields a 97% F1 score, which is high result while requiring a lesser amount
of time and complexity.

ACKNOWLEDGEMENTS
Authors thank the Department of Computer Science, College of Science, Mustansiriyah University,
Baghdad-Iraq for supporting this present work.

REFERENCES
[1] Z. El Khattabi, Y. Tabii, and A. Benkaddour, “Video shot boundary detection using the scale invariant feature transform and RGB
color channels.,” International Journal of Electrical & Computer Engineering (2088-8708), vol. 7, no. 5, 2017.
[2] L. Kong, “SIFT feature-based video camera boundary detection algorithm,” Complexity, vol. 2021, pp. 1–11, 2021.
[3] T. Zhou, F. Porikli, D. J. Crandall, L. Van Gool, and W. Wang, “A survey on deep learning technique for video segmentation,”
IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 45, no. 6, pp. 7099–7122, 2022.
[4] D. M. Thounaojam, T. Khelchandra, K. M. Singh, and S. Roy, “A genetic algorithm and fuzzy logic approach for video shot
boundary detection,” Computational intelligence and neuroscience, vol. 2016, 2016.
[5] E. Hato, “Temporal video segmentation using optical flow estimation,” Iraqi Journal of Science, pp. 4181–4194, 2021.
[6] H. Shao, Y. Qu, and W. Cui, “Shot boundary detection algorithm based on HSV histogram and HOG feature,” in 2015 International
Conference on Advanced Engineering Materials and Technology, Atlantis Press, pp. 951–957, 2015.
[7] S. Tippaya, S. Sitjongsataporn, T. Tan, M. M. Khan, and K. Chamnongthai, “Multi-modal visual features-based video shot boundary
detection,” IEEE Access, vol. 5, pp. 12563–12575, 2017, doi: 10.1109/ACCESS.2017.2717998.
[8] S. Akpinar and F. Alpaslan, “A novel optical flow-based representation for temporal video segmentation,” Turkish Journal of
Electrical Engineering and Computer Sciences, vol. 25, no. 5, pp. 3983–3993, 2017.
[9] M. Haroon, J. Baber, I. Ullah, S. M. Daudpota, M. Bakhtyar, and V. Devi, “Video scene detection using compact bag of visual word
models,” Advances in Multimedia, vol. 2018, pp. 1–9, 2018.
[10] F.-F. Duan and F. Meng, “Video shot boundary detection based on feature fusion and clustering technique,” IEEE Access, vol. 8,
pp. 214633–214645, 2020.
[11] Z. N. Idan, S. H. Abdulhussain, B. M. Mahmmod, K. A. Al-Utaibi, S. A. R. Al-Hadad, and S. M. Sait, “Fast shot boundary detection
based on separable moments and support vector machine,” IEEE Access, vol. 9, pp. 106412–106427, 2021.
[12] N. Kumar, “Shot boundary detection framework for video editing via adaptive thresholds and gradual curve point,” Turkish Journal
of Computer and Mathematics Education (TURCOMAT), vol. 12, no. 11, pp. 3820–3828, 2021.
[13] J. T. Jose, S. Rajkumar, M. R. Ghalib, A. Shankar, P. Sharma, and M. R. Khosravi, “Efficient shot boundary detection with multiple
visual representations,” Mobile Information Systems, vol. 2022, 2022.
[14] K. A. Akintoye, N. A. F. B. Ismial, N. Z. S. B. Othman, M. S. M. Rahim, and A. H. Abdullah, “Composite median Wiener filter
based technique for image enhancement.,” Journal of Theoretical & Applied Information Technology, vol. 96, no. 15, 2018.
[15] S. H. Majeed and N. A. M. Isa, “Adaptive entropy index histogram equalization for poor contrast images,” IEEE Access, vol. 9, pp.
6402–6437, 2020, doi: 10.1109/ACCESS.2020.3048148.
[16] A. M. Neto, A. C. Victorino, I. Fantoni, D. E. Zampieri, J. V. Ferreira, and D. A. Lima, “Image processing using Pearson’s
correlation coefficient: Applications on autonomous robotics,” in 2013 13th International Conference on Autonomous Robot
Systems, IEEE, pp. 1–6,2013.
[17] N. K. Ibrahim, A. H. Al-Saleh, and A. S. A. Jabar, “Texture and pixel intensity characterization-based image segmentation with
morphology and watershed techniques,” Indonesian Journal of Electrical Engineering and Computer Science, vol. 31, no. 3, pp.
1464–1477, 2023. doi: 10.11591/ijeecs.v31.i3.
[18] N. khalid, “Hybrid features of mask generated with gabor filter for texture analysis and sobel operator for image regions
segmentation using K-Means technique,” Journal La Multiapp, vol. 3, no. 5, pp. 250–258, 2022, doi:
10.37899/journallamultiapp.v3i5.743.
[19] X. Zheng, Q. Lei, R. Yao, Y. Gong, and Q. Yin, “Image segmentation based on adaptive K-means algorithm,” EURASIP Journal
on Image and Video Processing, vol. 2018, no. 1, pp. 1–10, 2018.
[20] U. Petronas, “Mean and standard deviation features of color histogram using laplacian filter for content-based image retrieval,”
Journal of Theoretical and Applied Information Technology, vol. 34, no. 1, pp. 1–7, 2011.
[21] R. Sammouda and A. El-Zaart, “An optimized approach for prostate image segmentation using K-means clustering algorithm with
elbow method,” Computational Intelligence and Neuroscience, vol. 2021, 2021.
[22] N. Dhanachandra and Y. J. Chanu, “A new approach of image segmentation method using K-means and kernel based subtractive
clustering methods,” International Journal of Applied Engineering Research, vol. 12, no. 20, pp. 10458–10464, 2017.
[23] N. M. Kwok, Q. P. Ha, and G. Fang, “Effect of color space on color image segmentation,” in 2009 2nd International Congress on
Image and Signal Processing, IEEE, pp. 1–5, 2009.
[24] L. David, “Distinctive image features from scale-invariant keypoints,” International journal of computer vision, vol. 60, pp. 91–
110, 2004.
[25] S. H. Abdulhussain, A. R. Ramli, M. I. Saripan, B. M. Mahmmod, S. A. R. Al-Haddad, and W. A. Jassim, “Methods and challenges
in shot boundary detection: a review,” Entropy, vol. 20, no. 4, p. 214, 2018.

Comput Sci Inf Technol, Vol. 5, No. 2, July 2024: 130-139

Comput Sci Inf Technol ISSN: 2722-3221  139

BIOGRAPHIES OF AUTHORS

Noor Khalid Ibrahim is lecturer at college of college of science, Mustansiriyah

University, Iraq. Received the B.Sc. degree in computer science from Department of
Computer, College of Science, Mustansiriyah University, Iraq. She holds a master degree in
computer science at 2015, with specialization in multi-media. Her research areas in image
processing. She can be contacted at email: [email protected].

Zinah Sadeq Abduljabbar is lectuter at collage of science, Mustansiriyah

university, Iraq. Received the B.Sc. degree in computer science from department of computer,
collage of science, Mustansiriyah university, Iraq. She holds a master degree in computer
science at 2014, with specialization in multi-media. she can be contacted at email:
[email protected].

Video shot boundary detection based on frames objects comparison and scale… (Noor Khalid Ibrahim)