Crimson Publishers Publish With Us Reprints e-Books Video articles

Full Text

COJ Robotics & Artificial Intelligence

Application of Data Fusion for Image Denoising - Basic Concepts

Jerzy Świątek and Krzysztof Brzostowski*

Department of Computer Science and Systems Engineering, Faculty of Computer Science and Management, Wroclaw University of Science and Technology, Poland

*Corresponding author: Krzysztof Brzostowski, Department of Computer Science and Systems Engineering, Faculty of Computer Science and Management, Wroclaw University of Science and Technology, Poland

Submission: August 7, 2021;Published: March 23, 2022

DOI: 10.31031/COJRA.2022.01.000524

ISSN:2832-4463
Volume1 Issue5

Abstract

In image processing, noise is introduced due to unperfected instrumentation and acquisition techniques. Noise in the original images degraded its quality and distorted useful information (e.g., medical images for diagnosis). The work proposes a noise reduction algorithm based on data fusion method to improve the quality of medical imaging based on computed tomography and magnetic resonance image scans.

Keywords: Data fusion; Artificial intelligence; Noise modelling; Image processing

Introduction

Data fusion includes theories, methods, techniques, and tools used to combine data from different sources to obtain a description of the observed object having an impact on, e.g., improving the quality of decision support or control [1]. The issue of data fusion is not a new concept. However, the development of data acquisition techniques with the use of measurement sensors, their processing methods, and computational equipment gives today new possibilities in this area [2]. A characteristic feature of data fusion is that the description of the observed object obtained with its use is more complete than the description based on algorithms processing data from different sources independently. Equally important is the fact that the use of data fusion tools reduces the influence of the uncertainty of the measurements used to describe the observed object. To make it possible, it is important to choose for measurement those quantities which are independent and complementary to each other. Data fusion may be understood as a set of methods and algorithms taken from different areas, which are selected to solve a specific task. Thus, when analysing task solutions based on data fusion, one may find methods and algorithms known from other areas, such as signal processing, image recognition, state estimation theory, or system identification. The most important issues considered in the area of data fusion include - apart from data fusion methods-also data acquisition techniques coming from different sources (e.g., various types of measurement sensors) and algorithms of data pre-processing. Among the most important premises which justify the use of data fusion methods and algorithms, we should mention the possibility to:

a. Improve measurement data quality (using appropriate algorithms allows to remove undesirable components from measurements),

b. Enhance the resilience and reliability of the designed system (redundancy guarantees the correct operation of the system in case of failure of some of its elements),

c. Improve measurement resolution (related to the reduction of quantification errors),

d. Shorten the measurement time (a group of various sensors enables measurements from different perspectives),

e. Reduce the ambiguity and uncertainty of measurements (by using data collected from different sources),

f. Increase the confidence (in many cases, a single sensor is not able to provide a sufficiently high level of confidence in the collected data),

g. Combining independent features and a priori knowledge (the possibility of building a complete description of the object, which considers the different perspectives captured by the measuring sensors embedded in it).

h. The purpose of data fusion is to combine data from different sources to build a more complete description of the observed object.

Case Presentation

The concept of data fusion can be also applied to solve the problem of image denoising. For example, CT (Computed Tomography) and MRI (Magnetic Resonance Image) are widely used in the diagnosis of various brain diseases like stroke, trauma, skull fracture, etc. Generally, MRI scans are preferred over CT due to better soft tissue differentiation and high contrast. It must be emphasised that in some circumstances CT imaging should be used as an alternative to MRI since its cost is lower, imaging time is shorter, good detection of bony details, etc. Unfortunately, CT images have also some drawbacks, e.g., radiation burden, low contrast, and artefacts due to spiral-off center [3,4]. Both CT and MRI images are burdened by noise. Noise is an component that enacts unwanted changes and modifications to the original image. In general, the noise feature leads to the unnecessary processing and utilization of resources for handling the burden data. In the literature there are agreements that the noise on CT scans is the additive Gaussian, whereas on MRI it is rician. To reduce the noise effects on images, some denoising filters can be used. One of the challenges to image denoisng is filtering out the unwanted components of the image without losing their important features. The task of unwanted components’ removal from acquired image can be formulated as follows

where y is the measurement (degraded source image), x is the true value of an image, v represents the undesirable component, and k=1,2,…,K . This expression is an accepted measurement model. The function of algorithms for removing unwanted components from measurements is to estimate

where

a x ̃(k) is the searched value after removal of undesirable components, and F is the algorithm for removing them.

We concern fusion of noisy CT and MRI scans to obtain an improved medical image. Let the CT and MRI images are decomposed into sub-bands by using wavelet transform [5,6]. In the results we obtain the set of wavelet coefficients for CT image and for MRI scan. Calculated coefficients W_CT and W_MRI are used in the next step to determine fused image, i.e., . The fusion rule in the considered approach is as follows

The last step of the proposed fusion scheme for CT and MRI image is to conduct the inverse wavelet transform on WFUSED to determine image with improved quality.

Discussion

The proposed in-the--the-work approach to denoise medical images utilizes the features of data fusion combined with CT and MRI scans. Our study shows that the suggested method is a promising direction of further work.

References

  1. Raol JR (2009) Multi-sensor data fusion with MATLAB, CRC press, USA.
  2. Sroka R (2002) Data fusion in measurement applications 48: 15-19.
  3. Li H, He X, Tao D, Tang Y, Wang R (2018) Joint medical image fusion, denoising and enhancement via discriminative low-rank sparse dictionary learning. Pattern Recognition 79: 130-146.
  4. Liu L, Xu L, Fang H (2020) Infrared and visible image fusion and denoising via ℓ2-ℓp norm minimization. Signal Processing 172: 107546.
  5. Diwakar M, Tripathi A, Joshi K, Sharma A, Singh P, et al. (2021) A comparative review: Medical image fusion using SWT and DWT. Materials Today: Proceedings 37(2): 3411-3416.
  6. Hermessi H, Mourali O, Zagrouba E (2021) Multimodal medical image fusion review: Theoretical background and recent advances. Signal Processing 183: 108036.

© 2022 Krzysztof Brzostowski. This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and build upon your work non-commercially.