2024 International Conference on Advances in Modern Age Technologies for Health and Engineering Science (AMATHE) | 979-8-3503-7156-7/24/$31.

00 ©2024 IEEE | DOI: 10.1109/AMATHE61652.2024.10582191

Effective Image Reconstruction using Various

Compressed Sensing Techniques
Mamidipaka Hema R.Gurunadha
Department of ECE Department of ECE
JNTUGVCEV JNTUGVCEV
Vizianagaram, Andhra Pradesh, India Vizianagaram, Andhra Pradesh, India
mhema.ece@jntugvcev.edu.in gururavva.ece@jntugvcev.edu.in

Jami Venkata Suman Madhavi Mallam

Department of ECE Department of ECE
GMR Institute of Technology PES Institute of Technology and Management
Rajam, Andhra Pradesh, India Shivamogga, Karnataka, India
venkatasuman.j@gmrit.edu.in gurumadhu432@gmail.com

Abstract—The prevailing information relies heavily on the I. INTRODUCTION

principles articulated by Shannon and Nyquist. These theories
assert that, to faithfully reconstruct a signal without distortion, Comprehensive sensing plays a crucial role in image
the sampling frequency must exceed twice the signal's maximum processing by providing a rich and detailed set of data about
frequency to prevent aliasing. Following this, signals undergo the scene or object being analyzed. This involves capturing
compression to eliminate inherent redundancies before information from various sources and modalities to create a
transmission over the channel. Despite the effectiveness of this holistic representation of the environment. The role of
approach, a significant drawback arises from the substantial comprehensive sensing in image processing can be summarized
processing overhead involved in sampling and compression. This
in several key aspects which Increased Information Capture by
limitation renders the scheme unsuitable for contemporary
applications, given the constraints of current computational
Comprehensive sensing involves the use of multiple sensors or
capabilities. Compressive Sensing (CS) introduces a novel imaging techniques to capture a wide range of information.
framework, grounded in signal decomposition and This results in more detailed and diverse data about the scene,
approximation theory. Serving as an alternative to the Nyquist enabling a more thorough analysis. Enhanced Feature
criteria, CS presents advantages such as reduced sensing time Extraction with different sensors can capture different features
and sampling rate. In this approach, signals are sampled below of an object or scene. By combining information from various
the Nyquist rate through linear projection onto a random basis, sources, comprehensive sensing allows for the extraction of a
ensuring exact reconstruction of the original signal. CS broader set of features. This is particularly beneficial for tasks
contributes to the reduction of power consumption and
such as object recognition, where multiple characteristics
computational complexity in handling digital data. The
extraction of information is facilitated by the utilization of a
contribute to accurate identification [1-3].
sensing matrix. In the context of image restoration, the CS
framework provides an innovative approach to recover high-
quality images from their compressed measurements. The success
of image restoration in compressive sensing relies on the
development of robust algorithms that can efficiently exploit
sparsity while addressing the inherent trade-off between
accuracy and computational complexity. We survey state-of-the-
art reconstruction algorithms, including iterative methods,
convex optimization, and deep learning approaches, showcasing
their strengths and limitations in restoring images from
compressed measurements. The success of image restoration in
compressive sensing relies on the development of robust Fig. 1. Model for comprehensive sensing
algorithms that can efficiently exploit sparsity while addressing Comprehensive sensing of an image as shown in figure 1.
the inherent trade-off between accuracy and computational Improved Accuracy and Robustness by utilizing data from
complexity. We survey state-of-the-art reconstruction algorithms,
multiple sensors, comprehensive sensing helps improve the
including iterative methods, convex optimization, and deep
learning approaches, showcasing their strengths and limitations accuracy and robustness of image processing algorithms.
in restoring images from compressed measurements. Redundant information can be leveraged to enhance reliability
and reduce the impact of noise or errors in individual sensor
Keywords— Wavelet domain, Signal processing algorithms, data. Multimodal Fusion with comprehensive sensing often
Transforms, reconstruction algorithm, convex optimization. involves the integration of data from various modalities, such
as visible light, infrared, radar, and others. Fusion of

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY BOMBAY. Downloaded on September 24,2024 at 15:48:00 UTC from IEEE Xplore. Restrictions apply.
 multimodal data allows for a more complete understanding of The flexibility in choosing sensing matrices and sampling
the scene, as each modality contributes unique information [4]. rates allows for optimization based on the requirements of the
Better Scene Understanding with the combination of specific imaging application. This adaptability can result in
information from different sensors enables a more improved reconstruction quality for images in applications like
comprehensive understanding of the scene. This is particularly medical imaging, surveillance, or remote sensing. This is
important in applications such as autonomous vehicles, beneficial in scenarios where rapid data acquisition and
surveillance systems, and medical imaging, where a thorough reconstruction are critical, such as in medical diagnostics or
understanding of the environment is essential. Adaptability to surveillance. By intelligently adapting sensing parameters,
Different Conditions with comprehensive sensing enables flexible compressed sensing can reduce the amount of data that
systems to adapt to different environmental conditions. For needs to be acquired and transmitted. This is advantageous in
example, by incorporating sensors with different wavelength applications where bandwidth or storage constraints are
sensitivities, imaging systems can operate effectively in significant factors. The adaptability of flexible compressed
varying lighting conditions. Advanced Imaging Applications in sensing can enhance the robustness of image reconstruction in
comprehensive sensing is essential for advanced imaging the presence of noise or artifacts. The ability to tailor the
applications, such as 3D reconstruction, augmented reality, and sensing process to specific features allows for more effective
virtual reality. These applications often require a wealth of data handling of challenging imaging conditions.
to create realistic and immersive visual experiences [5-6].

Fig. 3. Image transformation with wavelet

Fig. 2. Flexible compressed sensing tansformation Image transformation techniques in the context of
compressive sensing and wavelet shown in figure 3 involve
As seen in figure 2, Flexible compressed sensing (CS) is an applying transformations to an image to exploit its sparsity or
extension of traditional compressed sensing techniques that compressibility in a different domain, facilitating efficient
allows for more adaptability and customization based on compression and reconstruction from fewer measurements.
specific application requirements. In the context of image Here are a few image transformation techniques commonly
processing, flexible compressed sensing is designed to used in conjunction with compressive sensing. Wavelet
efficiently acquire and reconstruct images with reduced data Transform is widely used in compressive sensing applications.
acquisition and storage requirements. It decomposes an image into different frequency components,
Compressed sensing relies on the idea that many natural and often the resulting coefficients exhibit sparsity. Sparse
signals, including images, can be sparsely represented in a representation in the wavelet domain allows for efficient
certain domain (e.g., wavelet, Fourier, or sparse representation reconstruction from compressed measurements [7-9].
in some other transform). Flexible compressed sensing allows Sparse Coding and Dictionary Learning involves
for adaptability in choosing the representation basis, enabling representing an image as a linear combination of a few basis
better sparsity in specific applications. In traditional functions from a learned dictionary. Dictionary learning
compressed sensing, a random or predetermined sensing matrix algorithms aim to find an optimal set of basis functions that
is used during data acquisition. Flexible compressed sensing best represent the input data. Sparse coding and dictionary
allows for the adaptation of the sensing matrix based on the learning can enhance the sparsity of an image, facilitating
characteristics of the specific image or scene being captured. better compressive sensing performance.
This adaptability can enhance the efficiency of signal recovery. Flexible compressed sensing allows for the use of
Flexible compressed sensing enables the adjustment of customized reconstruction algorithms based on the
sampling rates for different regions or features within an image. characteristics of the specific sensing matrix and sampling
This adaptability is particularly useful in scenarios where strategy employed. This adaptability can lead to more efficient
certain parts of the image require higher fidelity or where and accurate image reconstruction. Overall, flexible
specific features need to be captured with greater detail. compressed sensing in image processing provides a versatile
framework that can be tailored to the specific needs and

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY BOMBAY. Downloaded on September 24,2024 at 15:48:00 UTC from IEEE Xplore. Restrictions apply.
 Original CS Sampling Scalar Entropy
Image Quantization Coder

Channel

Recovered CS Dequantization Entropy

Image Reconstruction Decoder
Fig. 4. Image transformation systems using CS

characteristics of different applications, offering advantages in overall understanding of the scene, providing a more complete
terms of efficiency, data reduction, and adaptability to diverse and holistic view.
imaging scenarios. Efficient Decision Making is detailed
information provided by comprehensive sensing supports Total Variation (TV) Minimization is a technique used for
more informed decision-making processes. Comprehensive regularization in image reconstruction problems. It aims to
sensing in image processing enhances the quality, reliability, preserve edges and details in images by penalizing the total
and scope of information available for analysis. This approach variation of pixel intensities. TV regularization is often
is particularly valuable in complex and dynamic environments employed in conjunction with compressive sensing to improve
where a thorough understanding of the scene is crucial for the quality of reconstructed images. Non-Local Means (NLM)
effective decision making. is a denoising technique that exploits similarities between non-
local image patches. It has been incorporated into compressive
II. MOTIVATION AND LITERATURE SURVEY
sensing frameworks to improve reconstruction quality by
The motivation for introducing comprehensive sensing in reducing noise in the sparse representations. In some cases,
image processing stems from the need to push the boundaries images or patches of images may exhibit low-rank structures.
of imaging capabilities, providing richer data for analysis, Low-rank matrix completion techniques are applied to recover
improving accuracy, and enabling applications in diverse and missing or compressed data efficiently. This can be particularly
challenging environments. This approach is essential for useful in scenarios where images have certain structural
advancing the capabilities of imaging systems across various dependencies.
domains. The introduction of comprehensive sensing in image Imaging systems often face changing environmental
processing is motivated by the desire to enhance the conditions. Comprehensive sensing allows for adaptability by
capabilities and performance of imaging systems across various incorporating sensors that operate optimally under different
applications. This approach aims to go beyond traditional conditions, such as low light, adverse weather, or varying
single-sensor imaging by incorporating data from multiple levels of visibility. The integration of comprehensive sensing is
sources and modalities. Here are several key motivations for essential for advanced imaging applications, including 3D
introducing comprehensive sensing in image processing. reconstruction, augmented reality, and virtual reality. These
Comprehensive sensing allows for the simultaneous applications require a wealth of data to create realistic and
acquisition of data from multiple sensors or imaging immersive visual experiences.
modalities. This results in a more comprehensive and detailed The motivation for image reconstruction using compressive
representation of the scene, capturing a broader range of sensing (CS) arises from the need to efficiently acquire and
information than traditional single-sensor approaches. By reconstruct images with reduced data, making it particularly
integrating data from diverse sensors, comprehensive sensing attractive in scenarios where data acquisition, storage, or
can improve the accuracy and reliability of image processing transmission resources are limited. The motivation for image
algorithms. Redundant information from different modalities reconstruction using compressive sensing lies in its ability to
can help mitigate errors and uncertainties associated with significantly reduce data requirements while maintaining the
individual sensors, leading to more robust and trustworthy quality of reconstructed images. This has broad implications
results [10]. across various fields, including remote sensing, medical
Different sensors are sensitive to different aspects of a scene imaging, communication, and resource-constrained
or object. Comprehensive sensing facilitates the extraction of a environments, where traditional imaging approaches may be
richer set of features, allowing for a more detailed and nuanced impractical or inefficient [11].
analysis. This is particularly beneficial for applications such as
object recognition, where a diverse set of features contributes III. DESIGN OF IMAGE TRANSFORMATION SYSTEM
to accurate identification. Comprehensive sensing enables the An image transformation system refers to a set of processes
fusion of data from various modalities, such as visible light, and components designed to modify or manipulate digital
infrared, radar, and more. This multimodal fusion enhances the images. These transformations can involve changing the
appearance, structure, or characteristics of the images for

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY BOMBAY. Downloaded on September 24,2024 at 15:48:00 UTC from IEEE Xplore. Restrictions apply.
 various purposes, such as enhancement, analysis, compression, surrounding areas. The goal is to intelligently select and adapt
or adaptation to specific applications. Image transformation patches from the available image data to seamlessly reconstruct
systems play a crucial role in image processing and computer missing regions. Reconstruction Algorithm carries the
vision. A compressed sensing image transformation system compressed measurements are then used in a reconstruction
shown in figure 4 is designed to efficiently acquire, process, algorithm to recover the original image. Common algorithms
and reconstruct images using the principles of compressed include basis pursuit, l1-minimization, and iterative methods.
sensing (CS). Compressed sensing is a signal processing Adaptive patch inpainting is a dynamic and evolving field,
technique that allows the reconstruction of sparse or with ongoing research to improve inpainting models, making
compressible signals from a reduced set of measurements. In
them more efficient and applicable to a broader range of image
the context of images, CS has been widely applied to reduce
types and inpainting scenarios. If you're interested in the latest
data acquisition and transmission requirements while
maintaining image quality. developments, consider looking into recent research papers and
publications in the field of image inpainting and restoration.
Images are often sparse or compressible in certain domains These algorithms exploit the sparsity or compressibility of
(e.g., wavelet, Fourier, or sparse representation) as shown in the image to reconstruct it accurately from the compressed
figure 5. The system begins by transforming the input image data. Adaptive Sampling carries compressed sensing image
into a sparse representation, emphasizing the most relevant transformation systems incorporate adaptive sampling
features. Compressed sensing involves acquiring a significantly strategies, adjusting the sampling pattern based on the
lower number of measurements or samples compared to the characteristics of the image. Adaptive sampling can further
Nyquist-Shannon sampling theorem. These measurements are improve reconstruction quality and efficiency.
obtained using a reduced set of sensors or measurements during
the data acquisition stage. The system employs a random IV. RESULT ANALYSIS
sensing matrix during the measurement process. This matrix is
typically a key component of CS algorithms and plays a crucial Image reconstruction using compressive sensing involves
role in capturing the essential information for image capturing a reduced set of measurements and then recovering
reconstruction. the original image from these measurements. Here's a general
outline of the process along with a Python example using the
cvxpy library for optimization. Image reconstruction using
compressive sensing typically involves the following steps, and
I'll outline a basic example using Python with NumPy and a
simple optimization solver as shown in figure 7.

Fig. 5. Compressed sensing image process

Encoding and Compression contains the acquired

measurements are encoded and compressed using CS
techniques. This may involve techniques such as random
sampling, and the measurements are often transformed into a
sparse representation using the sensing matrix. Transmission
Fig. 7. Origital image processing by CS
involves compressed data, representing the encoded image
information, can be transmitted efficiently through In this example:
communication channels with reduced bandwidth Step 1: We acquire a set of measurements by randomly
requirements. This is particularly beneficial for applications selecting a fraction of pixels from the original image.
with limited data transmission capabilities. Step 2: We create a sparse representation matrix (A), often
chosen based on the characteristics of the images. Here, we use
a simple identity matrix.
Step 3: We formulate an optimization problem to minimize the
L1-norm of the sparse coefficients (x) subject to the constraint
that the measurements match the acquired ones.
Step 4: We solve the optimization problem to obtain the sparse
coefficients.
Fig. 6. Adaptive Patch Inpainting Model
Step 5: We reconstruct the image from the sparse coefficients
and evaluate the quality of reconstruction using Peak Signal-to-
Adaptive Patch Inpainting shown in figure 6 refers to a Noise Ratio (PSNR).
technique used in image processing to fill in missing or Compressed sensing is a signal processing technique that
damaged parts of an image by extrapolating information from enables the reconstruction of sparse or compressible signals

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY BOMBAY. Downloaded on September 24,2024 at 15:48:00 UTC from IEEE Xplore. Restrictions apply.
 from a limited set of measurements. It has applications in Different transformations may be more suitable for certain
various fields, including image processing, medical imaging, types of images or applications. Additionally, the choice of
and communication. In the context of image reconstruction measurement matrices and optimization algorithms plays a
using compressed sensing, researchers explore methods to crucial role in achieving high-quality reconstructions.
efficiently reconstruct high-quality images from a reduced set Experimentation and evaluation using appropriate metrics, such
of measurements. This involves the design of suitable as PSNR or SSIM, are essential for assessing the effectiveness
measurement matrices, sparse representations, and optimization of the chosen techniques.
algorithms.
V. CONCLUSION
CS allows for the reconstruction of signals (such as
images) from a significantly reduced set of measurements
compared to traditional methods. This can be advantageous in
scenarios where data acquisition is costly or time-consuming.
Many natural images have sparse or compressible
representations in certain domains (e.g., wavelet or sparse
representation). CS leverages this sparsity, making it suitable
for applications where the signal has a sparse representation.
Compressive sensing is a versatile framework that can be
applied to various signal types, including images, audio, and
Fig. 8. Deep learning approach for image transmission
more. It has found applications in fields such as medical
imaging, remote sensing, and communication. Determining
Deep learning approaches shown in figure 8, particularly
whether image reconstruction using compressive sensing is the
convolutional neural networks (CNNs), have shown significant
"best" method compared to other approaches depends on
success in various image processing tasks, including image
various factors, including the specific characteristics of the
reconstruction. These networks can learn hierarchical features
images, the available resources, and the goals of the
directly from data and adapt to complex relationships in the
reconstruction.
input-output mapping. Residual networks (ResNets) have
Solving the optimization problem in compressive sensing
become popular in deep learning due to their ability to
can be computationally intensive, especially for large-scale
effectively train very deep networks. Multi-feature residual
images. The choice of optimization algorithm and the size of
networks likely involve the incorporation of multiple features
the problem can impact the computational efficiency. The
or information streams into the network architecture. This can
effectiveness of CS relies on the assumption that the signal of
enhance the model's ability to capture diverse aspects of the
interest has a sparse representation. In cases where this
input data.
assumption is not valid, other methods might perform better.
While CS can provide efficient reconstructions with fewer
measurements, the quality of reconstruction may vary
depending on the specific application and the level of
compression. The "best" method often depends on the specific
requirements of the application. For example, in medical
imaging, different reconstruction methods may be preferred
based on factors such as resolution, accuracy, and the
importance of real-time processing. With the rise of deep
learning, particularly convolutional neural networks (CNNs),
these approaches have shown remarkable success in image
reconstruction tasks. Deep learning methods can adaptively
Fig. 9. CNN Based image processing learn complex representations and patterns directly from the
data.
Recent research has explored the integration of deep Image reconstruction using compressive sensing is a
learning architectures with compressive sensing. Deep learning powerful and efficient approach, particularly when dealing with
models, such as convolutional neural networks (CNNs) as sparse signals. Its suitability depends on the specific
shown in figure 9, can automatically learn hierarchical features characteristics and constraints of the imaging scenario. In
and sparse representations for improved image reconstruction. recent years, the field has seen advancements in both
Combining sparse coding with low-rank matrix modeling has compressive sensing techniques and deep learning approaches
been explored for improved performance in compressive for image reconstruction.
sensing. This joint modeling can capture both sparse and
structured components in the data. REFERENCES
When implementing image transformation techniques with [ 1] M. Rani, S. B. Dhok and R. B. Deshmukh, "A Systematic Review of
compressive sensing, it's essential to consider the Compressive Sensing: Concepts, Implementations and Applications,"
IEEE Access, vol. 6, pp. 4875-4894, 2018.
characteristics of the specific images you are working with.

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY BOMBAY. Downloaded on September 24,2024 at 15:48:00 UTC from IEEE Xplore. Restrictions apply.
 [ 2] Y. Yunawan, I. Safitri and L. Novamizanti, "Compressive Sensing
for Image Watermarking Discrete Wavelet Transform and Spread
Spectrum", International Conference on Control Electronics
Renewable Energy and Communications (ICCEREC), pp. 99-103,
2018.
[ 3] Z. Gao, L. Dai, S. Han, C. -L. I, Z. Wang and L. Hanzo,
"Compressive Sensing Techniques for Next-Generation Wireless
Communications," IEEE Wireless Communications, vol. 25, no. 3,
pp. 144-153, June 2018.
[ 4] E. J. Candes and M. B. Wakin, "An Introduction to Compressive
Sampling", IEEE Signal Processing Magazine, pp. 21-30, 2008.
[ 5] B. Zhang, D. Xiao, M. Wang and J. Liang, "Privacy-Preserving
Compressed Sensing for Image Simultaneous Compression-
Encryption Applications," Data Compression Conference, Snowbird,
UT, USA, 2021, pp. 283-292.
[ 6] T. Prabhakar, G. Suresh, J .V. Suman, and N. V. Lalitha, “A
Comprehensive Methodological Analysis of Eigen Face Recognition
and Reconstruction”, International Journal of Computer Theory and
Engineering, vol. 4, no. 1, pp. 81-84, 2012.
[ 7] I. Fatwasauri and M. Rizkinia, "Compressive Sensing Image
Reconstruction With Total Variation and l2,1 Norm For Microwave
Imaging," 2019 IEEE International Conference on Innovative
Research and Development (ICIRD), Jakarta, Indonesia, 2019, pp. 1-
4.
[ 8] Z. Gao et al., "Structured Compressive Sensing Based Spatial-
Temporal Channel Estimation for FDD Massive MIMO", IEEE
Trans. Commun., vol. 64, no. 2, pp. 601-17, Feb. 2016.
[ 9] S. Mun and J. E. Fowler, "Block compressed sensing of images using
directional transforms", Proc. of IEEE Conference on Image
Processing, pp. 3021-3024, 2009.
[ 10] C. Zhao, J. Zhang, S. W. Ma and W. Gao, "Compressive-sensed image
coding via stripe-based DPCM", Proc. of Data Compression
Conference, pp. 171-180, 2016.
[ 11] D. L. Donoho, "Compressed sensing," IEEE Transactions on
Information Theory, vol. 52, no. 4, pp. 1289-1306, April 2006.

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY BOMBAY. Downloaded on September 24,2024 at 15:48:00 UTC from IEEE Xplore. Restrictions apply.