Image Restoration
Image Restoration
Submitted By
Harikesh Maurya (2008390100025)
[email protected]
Anand Rajvanshi (2008390100010)
[email protected]
Supervisor Name
Mr. Naveen Kumar Tiwari
Assistant Professor
2. Introduction
3. Problem Statement
4. Literature Review
5. Methodology
6. Code Implementation
8. Conclusion
9. Appendices
10. Reference
1
1 Abstract
Image restoration is a crucial task in computer vision aimed at recovering the original quality
and content of degraded images. It involves the removal of various types of distortions,
such as noise, blurring, and compression artifacts, which can degrade the visual quality and
interpretability of images. This abstract provides an overview of image restoration, including
its significance, key steps, and recent advancements.
The abstract then outlines the two main steps involved in image restoration: degradation
modeling and restoration algorithm design. Degradation modeling focuses on understanding
the causes of degradation and creating mathematical models to simulate them. It emphasizes
the importance of this step in guiding the design of effective restoration algorithms.
The abstract highlights the impact of deep learning-based methods on image restoration.
It explains the success of convolutional neural networks (CNNs) in tasks like denoising,
deblurring, and super-resolution. CNNs learn complex mappings between degraded and
clean images, enabling them to restore images with high visual quality and fidelity.
2 Introduction
The Telea algorithm is widely used in various applications such as image restoration, object
removal, and image editing. It has proven to be a valuable tool in the field of computer
vision, providing solutions for image inpainting tasks and contributing to the enhancement
and manipulation of visual content.
Telea algorithm is its ability to handle both small and large inpainting areas efficiently. It
also maintains the structure and texture of the original image, resulting in visually pleasing
inpainted results. However, it may struggle with complex textures or patterns and can
sometimes produce blurry or distorted regions if the inpainting area is too large or contains
intricate details.
inpainting algorithm is known as the Telea algorithm, named after its creator Alexandru
Telea. The Telea algorithm is a non-iterative, fast, and efficient method for inpainting
images. It utilizes a technique called ”fast marching” to propagate information from the
2
known parts of the image to the unknown regions, ultimately filling them with visually
coherent content.
Identify the damaged or missing regions:- The first step is to determine which parts of the
image need to be restored or filled in. These regions may be damaged due to scratches,
stains, or other forms of corruption, or they may be intentionally masked or removed.
Extract neighboring information:- The algorithm analyzes the surrounding areas of the dam-
aged regions to capture the texture, colors, and patterns. It looks for similar patches or
structures that can be used as a reference for inpainting.
Patch selection:- Based on the extracted information, the algorithm selects patches from the
undamaged regions that closely resemble the missing content. The selection process aims to
find patches that seamlessly blend with the surrounding areas.
Patch blending:- The selected patches are then blended or synthesized to generate new
content for the missing regions. Various techniques can be used, such as texture synthesis,
image blending, or mathematical optimization methods.
Refinement:- Inpainting algorithms often include refinement steps to improve the visual
quality and coherence of the restored regions. This may involve additional post-processing
techniques, such as edge-aware smoothing, color adjustment, or texture refinement.
3
3 Problem Statement
• Identify the damaged or missing regions.
• Patch selection.
• Patch blending.
• Refinement.
4
4 Literature Review
• ”Image Inpainting via Texture Synthesis” by Li-Yi Wei and Marc Levoy (2000) -
This paper introduced an approach to image inpainting based on texture synthesis.
It utilizes a non-parametric texture model to generate plausible textures in the
missing regions.
5
5 Methodology
In this filter, we define the size of kernel as 6,6(m,n). This kernel size describe
the neighbourhood over which the filter is applied. Next step is to the images
are traversed through each pixel. For each pixel, we have to calculate the average
value of intensities within the filtering kernel. The calculated intensity value
replaced with original pixel value. Repeat the calculation of average value step
and replace step for all pixel images. An adjustment is done of the window size
for our desired level of noise reduction and smoothing.
In this filter we define the size of kernel as 6,6(m,n). Next step is to traverse the
each pixel in the image. Now, we have to collect the pixel intensities for each
pixel within the filtering window. We have to calculate the median value after
sorting the collected pixel intensities. The calculated median is replace with the
original pixel value. Repeat the collecting of pixel intensity step and replacing it
with computed median step for all pixels in the image. An adjustment is done
of the window size which is based on our desired balance between preservation of
image details and noise reduction.
6
5.3 Histogram Equalization
A Gaussian filter is a low pass filter that commonly used for blurring and smooth-
ing the image. While preserving the important details, it reduce the noise. In
the code image path, kernel size and sigma are taken as input. The size of the
Gaussian are determined by the kernel size which determines the level of blurring
and smoothing. The standard deviation of the Gaussian distribution is controlled
by the sigma parameter which is used to generate the kernel.
6 Code Implementation
7
Figure 2: Implemented code part-1.
To implement this methodology for Speckle Damage reduction we have used CV2
and numpy for OpenCV and numerical operations respectively.
8
Figure 4: Mask image.
9
8 Conclusion
In conclusion, the inpaint Telea algorithm is a valuable tool for image restoration
and inpainting tasks. It offers a balance between efficiency and effectiveness,
producing satisfactory results in many scenarios. Nonetheless, it’s important to
consider the specific requirements and limitations of each inpainting task to choose
the most appropriate algorithm or technique.
9 Appendices
March-
a) Reviewing The Research Paper & Previous Work Done.
b) Collection of Dataset
April- Working on the model
May- Documentation on the completed model along with the model along with
the report.
10
References
11