Article Search
닫기

## Original Article

Split Viewer

International Journal of Fuzzy Logic and Intelligent Systems 2022; 22(4): 366-372

Published online December 25, 2022

https://doi.org/10.5391/IJFIS.2022.22.4.366

© The Korean Institute of Intelligent Systems

## MB-FL: Macro-Block Fuzzy Logic for Video Compression in Multimedia Applications

B. Veerasamy and C. M. Sangeetha

Department of Electronics and Communication Engineering, Hindusthan College of Engineering and Technology, Coimbatore, Tamil Nadu, India

Correspondence to :
B. Veerasamy (veerasamypapers2021@gmail.com)

Received: January 10, 2022; Revised: August 10, 2022; Accepted: December 11, 2022

This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/3.0/) which permits unrestricted noncommercial use, distribution, and reproduction in any medium, provided the original work is properly cited.

Methods to reduce the iterations needed to process a given video stream are referred to as video compression techniques. Videos consume a large amount of storage space on computational systems or handheld devices, and video compression is also used to lower the dimensionality of data owing to constraints on storage resources. In this study, we propose a macro block-based fuzzy logic video compression (MB-FL) algorithm. The proposed approach uses a fuzzy-based search to maintain pixel resolution, which is ideal for real time streaming media and thus increases peak signal-to-noise ratio (PSNR) and subjective quality. Compression is performed on repeated frames of data files via complex equations, and the repeating patterns are then substituted with smaller data or coding fragments. Owing bandwidth limitations, compression is often necessary to transmit and receive content over network connections. Using a fuzzy membership function, the multiscale aspect of our method evaluates the connection of individual components in the current frame to those in the reference frame. The results of an experimental evaluation show that the proposed approach significantly compressed files using a fuzzy-based search. We compared the performance of MB-FL with that of existing models to measure the quality of compressed video stream.

Keywords: Computational macro block based fuzzy logic video compression (MB-FL), Membership function (MF), Sum of absolute differences (SAD)

### 1. Introduction

Video compression methods are designed to reduce the size of data comprising video files while maintaining quality. Compression is a data encoding method that uses fewer bytes to enable more effective data processing and storage. Video compression is also referred to as video encoding, and is considered an important component of many application domains. Because video clips are relatively storage-intensive compared to e.g., text files, they tend to require a considerable amount of system memory to store and transmit. Decompression is the converse procedure, and the decoders can encrypt and decrypt compressed files. For example, an uncompressed movie comprises a massive volume of data, as shown by the following rough formula. With a size of 720 × 576 pixels (PAL), a refreshing frequency of 25 frames per second, and 8-bit color intensity, the corresponding throughput would have been required as given below.

720*576*8+2*(360*576*25*8)=1.66Mb/s(luminance+chrominance).

For high-definition television (HDTV), this value would be

1920*1080*60*8+2*(960*1080*60*8)=1.99Gb/s.

In addition, with powerful computer platforms, maintaining large volumes of data involves correspondingly intensive operational requirements. Furthermore, video streaming is suitable for compression, which can considerably reduce these issues. In particular, lossy image compression algorithms provide compression ratios for video sequences. Nevertheless, there is a trade-off between data volume (and, consequently processing time) and accuracy. The smaller a given file and the poorer the resolution, the greater the compression ratio. The encoding and decoding procedures also require computing resources. Figure 1 shows the compression and decompression processes involved in video compression. The original video was sent to the compressed video, which was then retrieved using a decompressor.

In this study, we propose a synthesis process based on a deep learning frame buffer augmentation method. The proposed approach provides state-of-the-art linearization outcomes and can create different pixels among both input data images to explore the impacts of omitting multiple information pixels while using their correlating interleaved pixels. Our methodology is quite adaptable, and may also be used with other frame-buffer interpolation techniques.

In many cases, deep video sequence formulation can effectively parallelize two adjacent frames, whereas video compression necessitates approximation across two remote images. To compact the digital information, lossless compression modules typically utilize a discrete cosine transform (DCT) transformation and linearization compression algorithm. DCT is calculated by multiplying two matrices as given below.

DBdct=DCT*DB*DCTt,

where DBdct is a transformed 16 × 16-pixels data block, DCT denotes is a pre-calculated DCT matrix, and DCTt is the transposed DCT matrix.

The contributions of this work are summarized as follows:

• - We proposed macro block-based fuzzy-logic video compression (MB-FL), which uses a fuzzy-based search to maintain pixel resolution.

• - The proposed approach is ideal for real-time streaming media as indicated by its increased visual quality and peak signal-to-noise ratio (PSNR).

• - This demonstrates that flexible pattern-matching yielded the best results while maintaining image resolution, which is ideal for real-time streaming media.

The remainder of this study is organized as follows. In this section, we have introduce our proposed method MB-FL for video compression. We briefly review the relevant literature in Section 2, and describe the proposed system in detail in Section 3. Section 4 presents our experimental results and some associate discussion. Section 5 concludes the work with some final remarks and suggests a promising avenue for future research.

### 2. Related Works

In recent years, compression algorithms for multimedia applications have been studied extensively to avoid degrading the quality of the compressed file. Compression involves several drawbacks as it reduces the quality of the image or video being compressed. In this section, we examine how compression techniques are utilized and methods of lossless compression. We also briefly highlight a few of the most important representative methods.

In 2021, Jalalpour et al. [1] suggested reducing the quantity of remaining information that requires encoding and proposed a hybrid compression algorithm approach that incorporated the benefits of residual encryption algorithms present in standard DCT-based compression algorithms to learn video sequence augmentation. For reasonable anchor-frame distances, this technique saves 5%–30% of the capacity while maintaining equal video quality, especially with VMAF (video multi-method assessment fusion). The efficiency of these technologies can be greatly improved by using DCT-based residual compression.

In 2010, Gorev and Ellervee [2] proposed a low-cost field-programmable gate array (FPGA) approach for original video encryption and broadcasting. For this reason, Bluetooth transmission was used as an efficient and available radio communication protocol. Because of its faster range, low energy consumption, and flexibility, the Bluetooth 2.0 protocol was adopted for connectivity. Byte-length encryption mechanisms must be improved to enhance this area. Parallelization of the quantized data must be altered to improve the compression ratio.

In 2011, Glaister et al. [3] developed a revolutionary compression algorithm pipeline that encoded video through targeted timestamp recognition and decoded it using this patch-based super-resolution enabling replay. Patch-based super-resolution methods discover identical patches among a high-resolution non-keyframes, as well as among the related frames. The size of the compressed file can be increased to minimize the amount of detail lost. Future research may be expected to focus on leveraging techniques to reducing computational resource requirements.

In 2011, Roy et al. [4] proposed a combined encryption and compaction method, and devised a simple yet powerful multilayer key streaming optimization methodology to encode video sequences. Their combined encryptor-encoder was nearly as quick as a standard H.264 compaction encoder. We frustrate code makers by extracting the multi-layer stream cipher used for encryption from the core video architecture itself. This also makes it easier to deal with transmission errors in real-time broadcasting. Systematic and scalable video coding (SVC) privacy is possible owing to this tiered technique.

In 2016, Al-Jawad et al. [5] developed an elevated compression technique for handheld devices with low storage space. They aimed to fine-tune a wavelet-based feature-preserving image lossless compression that was recently discovered for use on cellular telephones and PDAs. In the future, the concept of merely transferring statistical features rather than using a Huffman tree can be considered.

In 2013, Acharjee et al. [6] suggested an approach that discovered better movement estimates for greater mobility zones and required less time to compute a transformation matrix for lower motion zones. This technique is faster than existing block-matching algorithms, and the reconstructed images are of high standard. The resolution of the reconstructed image decreases when the block size is expanded; however, the overall process increases more rapidly, and the volume of the motion vector diminishes.

In 2011, Devi et al. [7] proposed diamond search (DS) as the most effective block-matching motion estimation (BMME) technique. A reworked DS algorithm altered the DS’s multiple search strategies. When compared to exhaustive and diamond searches, the modified diamond-square search (MDSS) approach achieves them by minimizing the overall search locations. The compression ratio can be improved using the MDSS algorithm and the PSNR ratios can be improved by upgrading the search strategy.

In 2010, Semertzidis et al. [8] proposed a real-time vision system for automatic vehicle tracking that utilized a network of autonomous tracking units (ATUs) to acquire and interpret images using one or more pre-calibrated cameras. The developed scheme was adaptable and suitable for a diverse range of products, such as vehicle tracking in highway tunnels and airport aircraft parking areas. An additional goal of this project was to review and validate various image-processing and management-fusion approaches before including them in the overall solution. They planned to expand the two additional systems in future work by further integrating automated traffic units as well as the delivery of highway traffic updates via the Internet to remote users.

The proposed system uses the macro-block fuzzy-logic approach (MB-FL) to compress the video without affecting its quality. As shown in Figure 2, the current and reference frames are blocked, which means that the specific frames are broken into different sub-frames or blocks.

The multi-resolution component estimates the connectivity of individual units in the current frame to those in the reference frame using a fuzzy membership function (MF). The fuzzy membership values for each macro-block are calculated using a Gaussian membership function (GMF). Fuzzy decision matching was performed according to the corresponding parameters, and the corresponding target block was chosen. Consequently, the motion matrices are determined with known coordinate frames, which can then be used to aid video compression.

The sum of absolute difference (SAD) is an evaluation of the common features between the partitions that are used as block sizes for comparison (8 × 8 or 16 × 16), the complete distinction across each pixel in the block of the reference frame, and the correlating pixel in the intended frame block. The critical component is the SAD, which is an evaluation of the common features between the blocks that are used as references are 8×8 or 16×16 blocks, with each pixel in the reference being distinguished in absolute terms. In contrast to other algorithms, the SAD is determined using the GMF, which is applied to the MB of the frame buffer and each pixel MB from the objective frame, producing a membership data matrix.

The still photos are preserved, and the I-frames are encoded individually, as shown in Figure 3. The truncated file shows all the necessary information because I-frames do not use inform decisions. They are encoded using a JPEG-like picture-compression algorithm. The change is “predicted” by the video transmitters from one frame to the next. The compression technique seems to have been more efficient when the estimates were closer. As a result, P-frames and B-frames are created. We begin by looking again for pixels with the lowest SAD compared to the desired frames. The focus point of the MF is determined mostly by pixels with the lowest SAD and minimal SAD. The search section has been built, and thus, the midsection of such a page has been modified, and a newer search effort is being built.

The definition of a residual image of size P×Q is a data matrix, where z(Piij) identifies the extent of connection of every image, as computed by the GMF. Each precise pixel score was considered once the GMF was created. Pi(i, j) obtains the value z(Piij), which is the appropriate membership level.

Piij=cat(PiXij,PiYij),Z(Piij)=e-(Piij-K)22η*σ2,

where PiX and PiY denote the residual images of the primary pictures X and Y, suffix ij reflects the picture’s component location, Piij is the amount of luminance of a picture’s residual image components, and η is the coefficient of enhancement. A standard deviation of a residual image is defined as σ, and it denotes the distance across the GMF. K denotes the core of the GMF, and it is referred to as the residual image average value.

σ=1P*Qi=1Pj=1Q(Pi(i,j)-K)2,K=1P*Qi=1Pj=1QPi(i,j),

where Pi(i, j) appears to be the image size of the image at location (i, j), whereas P and Q are the dimensions of the picture.

Both the present frame and the prior or existing set were used to compute the class labels mentioned in the preceding computations. Discrepancy membership values were used to determine the fuzzy conclusion for the expected picture sequences.

where zp(Piij) denotes membership in the current frame blocks and ze(Piij) denotes membership in the reference block. The defuzzification statement is the total of the absolute errors in membership values, including all units in which “M” stands for all macro units. The movement matrices of the sliding frames were calculated using the SAD. The alternatives for the two sequences are the units where the smallest SAD is attained.

### 4. Results and Discussion

The proposed methodology was applied in an experimental evaluation with the movie clip titled “tajmahal.mp4”. The panels were created using 352 × 288 pixels of an actual video stream. The movie data rate was 4 Mb/s. In the majority of previous studies, a monochrome or monochrome variant was analyzed, and the results, although in this study, a colorful panel extracted from actual footage, was used for evaluation.

Table 1 depicts the PSNR value between both the targeted panel and projected frames. When comparing the PSNR values of various strategies such as full search (FS), hexagonal search (HS), DS, the suggested MB-FL approach has a better PSNR value than most of the others.

FS, HS, DS, and the proposed MB-FL were tested with frame characteristics to determine their robustness. Our approach exhibited a better PSNR value than most of the others.

For an area and block size of 16, the predicted frame, residual frame, and vector with motion were analyzed. These frames are compared in Figure 4, which shows the proposed MB-FL model.

Tables 2, 3, and 4 show the PSNR, MSE, and SSIM comparisons (p = 4, b = 4), (p = 8, b = 8), and (p = 16, b = 16), respectively. When comparing the various strategies such as HS, DS FS, and the proposed approach, MB-FL exhibited better PSNR, MSE, and SSIM values than most of the others. Figure 5 shows a comparison in terms of PSNR, MSE and SSIM (p = 4, b = 4). Figure 6 shows the ((p = 8, b = 8) comparison table, and Figure 7 show the (p = 16, b = 16) comparison table.

### 5. Conclusion

The variations in the p and b factors and the PSNR score for the computation were assessed. For p = 4 and b = 4, the new MB-FL technique yielded a PSNR of 70.4968 relative to 65.4237, 69.3248, 67.5896 for the DS, HS, and FS algorithms, respectively. For p = 8, b = 8, p = 16, and b = 16, the suggested MB-FL search method produces better PSNR values of 69.8971 and 69.6287. This demonstrates that flexible pattern-matching yields the best results while maintaining image resolution, which is ideal for real-time streaming media. Future work may consider adopting various strategies and increasing PSNR values to increase the quality of compressed video.

### Conflict of Interest

Fig. 1.

Process of compression and decompression.

Fig. 2.

Proposed MB-FL model.

Fig. 3.

Representation of I-frame, B-frame, and P-frame in video compression.

Fig. 4.

(a) Predicted frame, (b) residual frame, (c) motion vector, (d) predicted frame MB-FL, (e) residual frame MB-FL, and (f) motion vector MB-FL with area p = 16, block size b = 16.

Fig. 5.

(p = 4, b = 4) PSNR, MSE and SSIM comparison.

Fig. 6.

(p = 8, b = 8) PSNR, MSE and SSIM comparison.

Fig. 7.

(p = 16, b = 16) PSNR, MSE and SSIM comparison.

Table. 1.

Table 1. PSNR comparison table (tajmahal.mp4).

Approachp = 4, b = 4p = 8, b = 8p = 16, b = 16
HS65.423766.785465.5284
DS69.324868.538768.1275
FS67.589667.687966.4691
Proposed MB-FL70.496869.897169.6287

Table. 2.

Table 2. Comparison of PSNR, MSE and SSIM (p = 4, b = 4).

ApproachPSNRMSESSIM
HS64.423740.5546.7
DS68.324855.7943.98
FS67.589657.3235.6
Proposed MB-FL70.496858.8929.5

Table. 3.

Table 3. Comparison of PSNR, MSE and SSIM (p = 8, b = 8).

ApproachPSNRMSESSIM
HS65.785439.7844.5
DS67.538740.8340.2
FS66.687955.3235.8
Proposed MB-FL69.897159.8928.5

Table. 4.

Table 4. Comparison of PSNR, MSE and SSIM (p = 16, b = 16).

ApproachPSNRMSESSIM
HS65.528444.546.7
DS68.127540.7943.6
FS65.469157.3235.3
Proposed MB-FL69.628759.8930.8

1. Jalalpour, Y, Wang, LY, Feng, WC, and Liu, F . FID: frame interpolation and DCT-based video compression., Proceedings of 2020 IEEE International Symposium on Multimedia (ISM, 2020, Naples, Italy, Array, pp.218-221. https://doi.org/10.1109/ISM.2020.00045
2. Gorev, M, and Ellervee, P . FPGA based system for video compression and transmission over Bluetooth., Proceedings of 2010 53rd IEEE International Midwest Symposium on Circuits and Systems, 2010, Seattle, WA, Array, pp.367-370. https://doi.org/10.1109/MWSCAS.2010.5548853
3. Glaister, J, Chan, C, Frankovich, M, Tang, A, and Wong, A . Hybrid video compression using selective keyframe identification and patch-based super-resolution., Proceedings of 2011 IEEE International Symposium on Multimedia, 2011, Dana Point, CA, Array, pp.105-110. https://doi.org/10.1109/ISM.2011.25
4. Roy, SD, Tian, J, Yu, H, and Zeng, W . A multi-layer key stream based approach for joint encryption and compression of H.264 video., Proceedings of 2011 IEEE International Conference on Multimedia and Expo, 2011, Barcelona, Spain, Array, pp.1-6. https://doi.org/10.1109/ICME.2011.6012179
5. Al-Jawad, N, Ehlers, J, and Jassim, S . An efficient realtime video compression algorithm with high feature preserving capability., Proceedings of SPIE, 2006, Array. article no. 625002
6. Acharjee, S, Biswas, D, Dey, N, Maji, P, and Chaudhuri, SS . An efficient motion estimation algorithm using division mechanism of low and high motion zone., Proceedings of 2013 International Multi-Conference on Automation, Computing, Communication, Control and Compressed Sensing (iMac4s), 2013, Kottayam, India, Array, pp.169-172. https://doi.org/10.1109/iMac4s.2013.6526402
7. Devi, AA, Sumalatha, MR, Priya, NM, Sukruthi, B, and Minisha, M . Modified diamond-square search technique for efficient motion estimation., Proceedings of 2011 International Conference on Recent Trends in Information Technology (ICRTIT), 2011, Chennai, India, Array, pp.1149-1153. https://doi.org/10.1109/ICRTIT.2011.5972450
8. Semertzidis, T, Dimitropoulos, K, Koutsia, A, and Grammalidis, N (2010). Video sensor network for real-time traffic monitoring and surveillance. IET Intelligent Transport Systems. 4, 103-112. https://doi.org/10.1049/iet-its.2008.0092

### Article

#### Original Article

International Journal of Fuzzy Logic and Intelligent Systems 2022; 22(4): 366-372

Published online December 25, 2022 https://doi.org/10.5391/IJFIS.2022.22.4.366

## MB-FL: Macro-Block Fuzzy Logic for Video Compression in Multimedia Applications

B. Veerasamy and C. M. Sangeetha

Department of Electronics and Communication Engineering, Hindusthan College of Engineering and Technology, Coimbatore, Tamil Nadu, India

Correspondence to:B. Veerasamy (veerasamypapers2021@gmail.com)

Received: January 10, 2022; Revised: August 10, 2022; Accepted: December 11, 2022

This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/3.0/) which permits unrestricted noncommercial use, distribution, and reproduction in any medium, provided the original work is properly cited.

### Abstract

Methods to reduce the iterations needed to process a given video stream are referred to as video compression techniques. Videos consume a large amount of storage space on computational systems or handheld devices, and video compression is also used to lower the dimensionality of data owing to constraints on storage resources. In this study, we propose a macro block-based fuzzy logic video compression (MB-FL) algorithm. The proposed approach uses a fuzzy-based search to maintain pixel resolution, which is ideal for real time streaming media and thus increases peak signal-to-noise ratio (PSNR) and subjective quality. Compression is performed on repeated frames of data files via complex equations, and the repeating patterns are then substituted with smaller data or coding fragments. Owing bandwidth limitations, compression is often necessary to transmit and receive content over network connections. Using a fuzzy membership function, the multiscale aspect of our method evaluates the connection of individual components in the current frame to those in the reference frame. The results of an experimental evaluation show that the proposed approach significantly compressed files using a fuzzy-based search. We compared the performance of MB-FL with that of existing models to measure the quality of compressed video stream.

Keywords: Computational macro block based fuzzy logic video compression (MB-FL), Membership function (MF), Sum of absolute differences (SAD)

### 1. Introduction

Video compression methods are designed to reduce the size of data comprising video files while maintaining quality. Compression is a data encoding method that uses fewer bytes to enable more effective data processing and storage. Video compression is also referred to as video encoding, and is considered an important component of many application domains. Because video clips are relatively storage-intensive compared to e.g., text files, they tend to require a considerable amount of system memory to store and transmit. Decompression is the converse procedure, and the decoders can encrypt and decrypt compressed files. For example, an uncompressed movie comprises a massive volume of data, as shown by the following rough formula. With a size of 720 × 576 pixels (PAL), a refreshing frequency of 25 frames per second, and 8-bit color intensity, the corresponding throughput would have been required as given below.

$720*576*8+2*(360*576*25*8)=1.66 Mb/s(luminance+chrominance).$

For high-definition television (HDTV), this value would be

$1920*1080*60*8+2*(960*1080*60*8)=1.99 Gb/s.$

In addition, with powerful computer platforms, maintaining large volumes of data involves correspondingly intensive operational requirements. Furthermore, video streaming is suitable for compression, which can considerably reduce these issues. In particular, lossy image compression algorithms provide compression ratios for video sequences. Nevertheless, there is a trade-off between data volume (and, consequently processing time) and accuracy. The smaller a given file and the poorer the resolution, the greater the compression ratio. The encoding and decoding procedures also require computing resources. Figure 1 shows the compression and decompression processes involved in video compression. The original video was sent to the compressed video, which was then retrieved using a decompressor.

In this study, we propose a synthesis process based on a deep learning frame buffer augmentation method. The proposed approach provides state-of-the-art linearization outcomes and can create different pixels among both input data images to explore the impacts of omitting multiple information pixels while using their correlating interleaved pixels. Our methodology is quite adaptable, and may also be used with other frame-buffer interpolation techniques.

In many cases, deep video sequence formulation can effectively parallelize two adjacent frames, whereas video compression necessitates approximation across two remote images. To compact the digital information, lossless compression modules typically utilize a discrete cosine transform (DCT) transformation and linearization compression algorithm. DCT is calculated by multiplying two matrices as given below.

$DBdct=DCT*DB*DCTt,$

where DBdct is a transformed 16 × 16-pixels data block, DCT denotes is a pre-calculated DCT matrix, and DCTt is the transposed DCT matrix.

The contributions of this work are summarized as follows:

• - We proposed macro block-based fuzzy-logic video compression (MB-FL), which uses a fuzzy-based search to maintain pixel resolution.

• - The proposed approach is ideal for real-time streaming media as indicated by its increased visual quality and peak signal-to-noise ratio (PSNR).

• - This demonstrates that flexible pattern-matching yielded the best results while maintaining image resolution, which is ideal for real-time streaming media.

The remainder of this study is organized as follows. In this section, we have introduce our proposed method MB-FL for video compression. We briefly review the relevant literature in Section 2, and describe the proposed system in detail in Section 3. Section 4 presents our experimental results and some associate discussion. Section 5 concludes the work with some final remarks and suggests a promising avenue for future research.

### 2. Related Works

In recent years, compression algorithms for multimedia applications have been studied extensively to avoid degrading the quality of the compressed file. Compression involves several drawbacks as it reduces the quality of the image or video being compressed. In this section, we examine how compression techniques are utilized and methods of lossless compression. We also briefly highlight a few of the most important representative methods.

In 2021, Jalalpour et al. [1] suggested reducing the quantity of remaining information that requires encoding and proposed a hybrid compression algorithm approach that incorporated the benefits of residual encryption algorithms present in standard DCT-based compression algorithms to learn video sequence augmentation. For reasonable anchor-frame distances, this technique saves 5%–30% of the capacity while maintaining equal video quality, especially with VMAF (video multi-method assessment fusion). The efficiency of these technologies can be greatly improved by using DCT-based residual compression.

In 2010, Gorev and Ellervee [2] proposed a low-cost field-programmable gate array (FPGA) approach for original video encryption and broadcasting. For this reason, Bluetooth transmission was used as an efficient and available radio communication protocol. Because of its faster range, low energy consumption, and flexibility, the Bluetooth 2.0 protocol was adopted for connectivity. Byte-length encryption mechanisms must be improved to enhance this area. Parallelization of the quantized data must be altered to improve the compression ratio.

In 2011, Glaister et al. [3] developed a revolutionary compression algorithm pipeline that encoded video through targeted timestamp recognition and decoded it using this patch-based super-resolution enabling replay. Patch-based super-resolution methods discover identical patches among a high-resolution non-keyframes, as well as among the related frames. The size of the compressed file can be increased to minimize the amount of detail lost. Future research may be expected to focus on leveraging techniques to reducing computational resource requirements.

In 2011, Roy et al. [4] proposed a combined encryption and compaction method, and devised a simple yet powerful multilayer key streaming optimization methodology to encode video sequences. Their combined encryptor-encoder was nearly as quick as a standard H.264 compaction encoder. We frustrate code makers by extracting the multi-layer stream cipher used for encryption from the core video architecture itself. This also makes it easier to deal with transmission errors in real-time broadcasting. Systematic and scalable video coding (SVC) privacy is possible owing to this tiered technique.

In 2016, Al-Jawad et al. [5] developed an elevated compression technique for handheld devices with low storage space. They aimed to fine-tune a wavelet-based feature-preserving image lossless compression that was recently discovered for use on cellular telephones and PDAs. In the future, the concept of merely transferring statistical features rather than using a Huffman tree can be considered.

In 2013, Acharjee et al. [6] suggested an approach that discovered better movement estimates for greater mobility zones and required less time to compute a transformation matrix for lower motion zones. This technique is faster than existing block-matching algorithms, and the reconstructed images are of high standard. The resolution of the reconstructed image decreases when the block size is expanded; however, the overall process increases more rapidly, and the volume of the motion vector diminishes.

In 2011, Devi et al. [7] proposed diamond search (DS) as the most effective block-matching motion estimation (BMME) technique. A reworked DS algorithm altered the DS’s multiple search strategies. When compared to exhaustive and diamond searches, the modified diamond-square search (MDSS) approach achieves them by minimizing the overall search locations. The compression ratio can be improved using the MDSS algorithm and the PSNR ratios can be improved by upgrading the search strategy.

In 2010, Semertzidis et al. [8] proposed a real-time vision system for automatic vehicle tracking that utilized a network of autonomous tracking units (ATUs) to acquire and interpret images using one or more pre-calibrated cameras. The developed scheme was adaptable and suitable for a diverse range of products, such as vehicle tracking in highway tunnels and airport aircraft parking areas. An additional goal of this project was to review and validate various image-processing and management-fusion approaches before including them in the overall solution. They planned to expand the two additional systems in future work by further integrating automated traffic units as well as the delivery of highway traffic updates via the Internet to remote users.

### 3. MB-FL

The proposed system uses the macro-block fuzzy-logic approach (MB-FL) to compress the video without affecting its quality. As shown in Figure 2, the current and reference frames are blocked, which means that the specific frames are broken into different sub-frames or blocks.

The multi-resolution component estimates the connectivity of individual units in the current frame to those in the reference frame using a fuzzy membership function (MF). The fuzzy membership values for each macro-block are calculated using a Gaussian membership function (GMF). Fuzzy decision matching was performed according to the corresponding parameters, and the corresponding target block was chosen. Consequently, the motion matrices are determined with known coordinate frames, which can then be used to aid video compression.

The sum of absolute difference (SAD) is an evaluation of the common features between the partitions that are used as block sizes for comparison (8 × 8 or 16 × 16), the complete distinction across each pixel in the block of the reference frame, and the correlating pixel in the intended frame block. The critical component is the SAD, which is an evaluation of the common features between the blocks that are used as references are 8×8 or 16×16 blocks, with each pixel in the reference being distinguished in absolute terms. In contrast to other algorithms, the SAD is determined using the GMF, which is applied to the MB of the frame buffer and each pixel MB from the objective frame, producing a membership data matrix.

The still photos are preserved, and the I-frames are encoded individually, as shown in Figure 3. The truncated file shows all the necessary information because I-frames do not use inform decisions. They are encoded using a JPEG-like picture-compression algorithm. The change is “predicted” by the video transmitters from one frame to the next. The compression technique seems to have been more efficient when the estimates were closer. As a result, P-frames and B-frames are created. We begin by looking again for pixels with the lowest SAD compared to the desired frames. The focus point of the MF is determined mostly by pixels with the lowest SAD and minimal SAD. The search section has been built, and thus, the midsection of such a page has been modified, and a newer search effort is being built.

The definition of a residual image of size P×Q is a data matrix, where z(Piij) identifies the extent of connection of every image, as computed by the GMF. Each precise pixel score was considered once the GMF was created. Pi(i, j) obtains the value z(Piij), which is the appropriate membership level.

$Piij=cat(PiXij,PiYij),Z(Piij)=e-(Piij-K)22η*σ2,$

where PiX and PiY denote the residual images of the primary pictures X and Y, suffix ij reflects the picture’s component location, Piij is the amount of luminance of a picture’s residual image components, and η is the coefficient of enhancement. A standard deviation of a residual image is defined as σ, and it denotes the distance across the GMF. K denotes the core of the GMF, and it is referred to as the residual image average value.

$σ=1P*Q∑i=1P∑j=1Q(Pi(i,j)-K)2,K=1P*Q∑i=1P∑j=1QPi(i,j),$

where Pi(i, j) appears to be the image size of the image at location (i, j), whereas P and Q are the dimensions of the picture.

Both the present frame and the prior or existing set were used to compute the class labels mentioned in the preceding computations. Discrepancy membership values were used to determine the fuzzy conclusion for the expected picture sequences.

$ef=(abs(ze(Piij)-zp(Piij)),SAD=∑l=1Mef,$

where zp(Piij) denotes membership in the current frame blocks and ze(Piij) denotes membership in the reference block. The defuzzification statement is the total of the absolute errors in membership values, including all units in which “M” stands for all macro units. The movement matrices of the sliding frames were calculated using the SAD. The alternatives for the two sequences are the units where the smallest SAD is attained.

### 4. Results and Discussion

The proposed methodology was applied in an experimental evaluation with the movie clip titled “tajmahal.mp4”. The panels were created using 352 × 288 pixels of an actual video stream. The movie data rate was 4 Mb/s. In the majority of previous studies, a monochrome or monochrome variant was analyzed, and the results, although in this study, a colorful panel extracted from actual footage, was used for evaluation.

Table 1 depicts the PSNR value between both the targeted panel and projected frames. When comparing the PSNR values of various strategies such as full search (FS), hexagonal search (HS), DS, the suggested MB-FL approach has a better PSNR value than most of the others.

FS, HS, DS, and the proposed MB-FL were tested with frame characteristics to determine their robustness. Our approach exhibited a better PSNR value than most of the others.

For an area and block size of 16, the predicted frame, residual frame, and vector with motion were analyzed. These frames are compared in Figure 4, which shows the proposed MB-FL model.

Tables 2, 3, and 4 show the PSNR, MSE, and SSIM comparisons (p = 4, b = 4), (p = 8, b = 8), and (p = 16, b = 16), respectively. When comparing the various strategies such as HS, DS FS, and the proposed approach, MB-FL exhibited better PSNR, MSE, and SSIM values than most of the others. Figure 5 shows a comparison in terms of PSNR, MSE and SSIM (p = 4, b = 4). Figure 6 shows the ((p = 8, b = 8) comparison table, and Figure 7 show the (p = 16, b = 16) comparison table.

### 5. Conclusion

The variations in the p and b factors and the PSNR score for the computation were assessed. For p = 4 and b = 4, the new MB-FL technique yielded a PSNR of 70.4968 relative to 65.4237, 69.3248, 67.5896 for the DS, HS, and FS algorithms, respectively. For p = 8, b = 8, p = 16, and b = 16, the suggested MB-FL search method produces better PSNR values of 69.8971 and 69.6287. This demonstrates that flexible pattern-matching yields the best results while maintaining image resolution, which is ideal for real-time streaming media. Future work may consider adopting various strategies and increasing PSNR values to increase the quality of compressed video.

### Fig 1.

Figure 1.

Process of compression and decompression.

The International Journal of Fuzzy Logic and Intelligent Systems 2022; 22: 366-372https://doi.org/10.5391/IJFIS.2022.22.4.366

### Fig 2.

Figure 2.

Proposed MB-FL model.

The International Journal of Fuzzy Logic and Intelligent Systems 2022; 22: 366-372https://doi.org/10.5391/IJFIS.2022.22.4.366

### Fig 3.

Figure 3.

Representation of I-frame, B-frame, and P-frame in video compression.

The International Journal of Fuzzy Logic and Intelligent Systems 2022; 22: 366-372https://doi.org/10.5391/IJFIS.2022.22.4.366

### Fig 4.

Figure 4.

(a) Predicted frame, (b) residual frame, (c) motion vector, (d) predicted frame MB-FL, (e) residual frame MB-FL, and (f) motion vector MB-FL with area p = 16, block size b = 16.

The International Journal of Fuzzy Logic and Intelligent Systems 2022; 22: 366-372https://doi.org/10.5391/IJFIS.2022.22.4.366

### Fig 5.

Figure 5.

(p = 4, b = 4) PSNR, MSE and SSIM comparison.

The International Journal of Fuzzy Logic and Intelligent Systems 2022; 22: 366-372https://doi.org/10.5391/IJFIS.2022.22.4.366

### Fig 6.

Figure 6.

(p = 8, b = 8) PSNR, MSE and SSIM comparison.

The International Journal of Fuzzy Logic and Intelligent Systems 2022; 22: 366-372https://doi.org/10.5391/IJFIS.2022.22.4.366

### Fig 7.

Figure 7.

(p = 16, b = 16) PSNR, MSE and SSIM comparison.

The International Journal of Fuzzy Logic and Intelligent Systems 2022; 22: 366-372https://doi.org/10.5391/IJFIS.2022.22.4.366

PSNR comparison table (tajmahal.mp4).

Approachp = 4, b = 4p = 8, b = 8p = 16, b = 16
HS65.423766.785465.5284
DS69.324868.538768.1275
FS67.589667.687966.4691
Proposed MB-FL70.496869.897169.6287

Comparison of PSNR, MSE and SSIM (p = 4, b = 4).

ApproachPSNRMSESSIM
HS64.423740.5546.7
DS68.324855.7943.98
FS67.589657.3235.6
Proposed MB-FL70.496858.8929.5

Comparison of PSNR, MSE and SSIM (p = 8, b = 8).

ApproachPSNRMSESSIM
HS65.785439.7844.5
DS67.538740.8340.2
FS66.687955.3235.8
Proposed MB-FL69.897159.8928.5

Comparison of PSNR, MSE and SSIM (p = 16, b = 16).

ApproachPSNRMSESSIM
HS65.528444.546.7
DS68.127540.7943.6
FS65.469157.3235.3
Proposed MB-FL69.628759.8930.8

### References

1. Jalalpour, Y, Wang, LY, Feng, WC, and Liu, F . FID: frame interpolation and DCT-based video compression., Proceedings of 2020 IEEE International Symposium on Multimedia (ISM, 2020, Naples, Italy, Array, pp.218-221. https://doi.org/10.1109/ISM.2020.00045
2. Gorev, M, and Ellervee, P . FPGA based system for video compression and transmission over Bluetooth., Proceedings of 2010 53rd IEEE International Midwest Symposium on Circuits and Systems, 2010, Seattle, WA, Array, pp.367-370. https://doi.org/10.1109/MWSCAS.2010.5548853
3. Glaister, J, Chan, C, Frankovich, M, Tang, A, and Wong, A . Hybrid video compression using selective keyframe identification and patch-based super-resolution., Proceedings of 2011 IEEE International Symposium on Multimedia, 2011, Dana Point, CA, Array, pp.105-110. https://doi.org/10.1109/ISM.2011.25
4. Roy, SD, Tian, J, Yu, H, and Zeng, W . A multi-layer key stream based approach for joint encryption and compression of H.264 video., Proceedings of 2011 IEEE International Conference on Multimedia and Expo, 2011, Barcelona, Spain, Array, pp.1-6. https://doi.org/10.1109/ICME.2011.6012179
5. Al-Jawad, N, Ehlers, J, and Jassim, S . An efficient realtime video compression algorithm with high feature preserving capability., Proceedings of SPIE, 2006, Array. article no. 625002
6. Acharjee, S, Biswas, D, Dey, N, Maji, P, and Chaudhuri, SS . An efficient motion estimation algorithm using division mechanism of low and high motion zone., Proceedings of 2013 International Multi-Conference on Automation, Computing, Communication, Control and Compressed Sensing (iMac4s), 2013, Kottayam, India, Array, pp.169-172. https://doi.org/10.1109/iMac4s.2013.6526402
7. Devi, AA, Sumalatha, MR, Priya, NM, Sukruthi, B, and Minisha, M . Modified diamond-square search technique for efficient motion estimation., Proceedings of 2011 International Conference on Recent Trends in Information Technology (ICRTIT), 2011, Chennai, India, Array, pp.1149-1153. https://doi.org/10.1109/ICRTIT.2011.5972450
8. Semertzidis, T, Dimitropoulos, K, Koutsia, A, and Grammalidis, N (2010). Video sensor network for real-time traffic monitoring and surveillance. IET Intelligent Transport Systems. 4, 103-112. https://doi.org/10.1049/iet-its.2008.0092