Q1. Compare the performance of JPEG and PNG compression methods.

JPEG (Joint Photographic Experts

Aspect PNG (Portable Network Graphics)
Group)

Compression Lossy compression, leading to some Lossless compression, retaining 100%

Type quality loss during compression. of the original quality.

Uses Discrete Cosine Transform (DCT) to Uses Deflate compression (a

Compression
compress image data by reducing less combination of LZ77 and Huffman
Algorithm
noticeable details. coding) to reduce redundancy.

Moderate compression ratio (2:1 to

Compression High compression ratio (10:1 to 20:1),
10:1), resulting in larger files compared
Ratio achieving significantly smaller file sizes.
to JPEG.

Quality is reduced, especially at higher Preserves original quality with no

Image Quality
compression levels (e.g., block artifacts). degradation.

Ideal for photographic images with Suitable for images with sharp edges,
Best Use Case
gradients and complex color variations. text, and graphics like logos.

Smaller file sizes due to lossy Larger file sizes because of lossless
File Size
compression. compression.

Transparency Does not support transparency (alpha Fully supports transparency through an
Support channel). alpha channel.

Supports up to 24-bit color (16.7 million Supports up to 48-bit color and

Color Depth
colors). grayscale with 16-bit depth.

Computational Lower computational cost, faster Higher computational cost, slower than
Demand encoding and decoding. JPEG.

Logos, icons, web graphics, and

Photography, web images, social media,
Applications applications requiring high-quality
and email sharing.
visuals.

Quality Loss in Quality degrades with repeated saves No quality loss even with repeated
Edits (generation loss). saves.
Q2. Explain the concepts of Dilation and erosion in morphological image processing.

Ans

Dilation and Erosion in Morphological Image Processing

Morphological image processing is a set of image processing operations that process images based
on shapes. Two fundamental operations in this domain are dilation and erosion, which are widely
used in image segmentation, boundary detection, and noise removal.

1. Dilation

Dilation is a morphological operation that expands the boundaries of objects in a binary image. It
adds pixels to the edges of objects, increasing their size.

• How It Works:

o A structuring element (a predefined shape such as a square, disk, or cross) is placed

over the image.

o If at least one pixel in the neighborhood (defined by the structuring element) is part
of the object, the output pixel becomes part of the object (set to 1).

o This process "grows" or "dilates" the object.

• Effect:

o Increases the size of bright regions (foreground).

o Fills small holes or gaps in the object.

• Formula:

A⊕B={z∣(Bz∩A)≠∅}A \oplus B = \{ z | (B_z \cap A) \neq \emptyset \}A⊕B={z∣(Bz∩A) =∅}

Where AAA is the input image, BBB is the structuring element, and zzz is the translation vector.

• Applications:

o Bridging gaps between object parts.

o Filling small holes in binary images.

o Highlighting boundaries of objects.

2. Erosion

Erosion is a morphological operation that shrinks the boundaries of objects in a binary image. It
removes pixels on object boundaries, reducing their size.

• How It Works:

o A structuring element is placed over the image.

 o If all the pixels in the neighborhood (defined by the structuring element) are part of
the object, the output pixel remains part of the object (set to 1); otherwise, it is
removed (set to 0).

o This process "erodes" or "shrinks" the object.

• Effect:

o Reduces the size of bright regions (foreground).

o Removes small objects or noise.

• Formula:

A⊖B={z∣Bz⊆A}A \ominus B = \{ z | B_z \subseteq A \}A⊖B={z∣Bz⊆A}

Where AAA is the input image, BBB is the structuring element, and zzz is the translation vector.

• Applications:

o Removing noise.

o Separating objects that are close to each other.

o Reducing the thickness of objects.

Q3. Explain Image restoration and challenges associated with it.

Ans-

Image Restoration

Image restoration is the process of reconstructing or recovering an image that has been degraded by
various factors, such as noise, blur, or distortion, to achieve a version that is as close as possible to
the original image. It aims to reverse the degradation process, often requiring mathematical models
and algorithms.

Key Objectives of Image Restoration

1. Remove Noise: Eliminate random variations in brightness or color.

2. Deblur Images: Restore sharpness lost due to motion, defocusing, or atmospheric

conditions.

3. Correct Distortions: Address issues like geometric distortions caused by camera lenses.

4. Enhance Details: Recover fine details obscured due to degradation.

Common Degradation Factors

1. Noise:

o Types: Gaussian noise, salt-and-pepper noise, speckle noise.

 2. Blur:

o Types: Motion blur, out-of-focus blur.

3. Geometric Distortion:

o Caused by lens imperfections or perspective shifts.

4. Environmental Effects:

o Atmospheric conditions like fog, rain, or heat waves.

5. Compression Artifacts:

o Loss of quality due to lossy compression techniques.

Techniques in Image Restoration

1. Noise Reduction:

o Filters: Mean filter, Median filter, Wiener filter.

o Advanced methods: Wavelet denoising, Non-local means filtering.

2. Deblurring:

o Methods: Inverse filtering, Wiener filtering, Blind deconvolution.

3. Geometric Corrections:

o Techniques: Affine transformations, warping algorithms.

4. Model-Based Restoration:

o Using degradation models like convolution models to restore images:

g(x,y)=f(x,y)∗h(x,y)+n(x,y)g(x, y) = f(x, y) \ast h(x, y) + n(x, y)g(x,y)=f(x,y)∗h(x,y)+n(x,y)
Where:

▪ g(x,y)g(x, y)g(x,y): Degraded image.

▪ f(x,y)f(x, y)f(x,y): Original image.

▪ h(x,y)h(x, y)h(x,y): Point Spread Function (PSF) describing blur.

▪ n(x,y)n(x, y)n(x,y): Noise.

5. Machine Learning and AI:

o Neural networks and deep learning models are now widely used for tasks like super-
resolution and removing noise/artifacts.

Challenges in Image Restoration

1. Modeling the Degradation Process:

 o Accurately estimating the degradation function (e.g., blur kernel or noise model) is
complex and application-specific.

2. Noise Complexity:

o Real-world noise often deviates from ideal models like Gaussian or uniform noise,
making it difficult to remove.

3. Under-Defined Problems:

o Restoration problems are often ill-posed, meaning there might not be a unique or
stable solution.

4. Trade-Off Between Detail and Artifacts:

o Excessive restoration can introduce artifacts, while conservative restoration may

leave degradation uncorrected.

5. Computational Complexity:

o High-quality restoration techniques, especially AI-based methods, often require

significant computational resources.

Q4. Explain the use of region growing algorithm for region based image Segmentation.

Ans-

Region Growing Algorithm for Region-Based Image Segmentation

Region growing is a fundamental technique in region-based image segmentation, used to divide an

image into meaningful regions by grouping pixels with similar properties. This algorithm starts with a
seed point and grows a region by adding neighboring pixels that satisfy predefined criteria (e.g.,
intensity, color, or texture similarity).

How Region Growing Works

1. Initialization:

o Select one or more seed points as the starting locations for segmentation. These
points are chosen based on prior knowledge or criteria like intensity values.

2. Growing Criteria:
 o Define a set of conditions that determine whether a neighboring pixel should be
added to the region. Common criteria include:

▪ Pixel intensity similarity.

▪ Euclidean distance in the feature space (e.g., RGB color space).

▪ Gradient magnitude threshold.

3. Region Growth:

o Iteratively add neighboring pixels to the region if they meet the growing criteria.

o Update the region's properties (e.g., mean intensity) after adding new pixels.

4. Termination:

o Stop growing when no more pixels satisfy the criteria or when a predefined stopping
condition (e.g., region size) is met.

Steps of the Algorithm

1. Input:

o An image III, seed points SSS, and a similarity threshold TTT.

2. Region Initialization:

o Start with the seed point (x,y)(x, y)(x,y), and initialize the region RRR with SSS.

3. Iterative Growth:

o For each pixel ppp in RRR:

▪ Check its neighbors N(p)N(p)N(p).

▪ Add q∈N(p)q \in N(p)q∈N(p) to RRR if: ∣I(q)−I(p)∣≤T|I(q) - I(p)| \leq

T∣I(q)−I(p)∣≤T

▪ Update the region mean or other properties if qqq is added.

4. Output:

o A segmented region RRR that satisfies the criteria.

Advantages of Region Growing

1. Simplicity:

o Easy to understand and implement.

2. Connectivity:

o Ensures that segmented regions are connected.

3. Flexible Criteria:
 o Can be tailored to specific applications by adjusting the similarity criteria.

4. Good Results for Homogeneous Regions:

o Performs well when the regions have clear and distinct properties.