Scribd
Upload
43 views

Image Colorization Using AI

Uploaded by

qccqstha
Copyright
© © All Rights Reserved
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
43 views

Image Colorization Using AI

Uploaded by

qccqstha
Copyright
© © All Rights Reserved
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
You are on page 1/ 19

Image Colorization

Using Deep Learning


introduction

• Colorization, the process of adding color to grayscale


images, has gained significant attention in recent
years.
• However, accurately and realistically colorizing
grayscale images remains a challenging task due to
the inherent ambiguity in determining the original
colors.
• This project aims to develop a robust colorization
model capable of producing high-quality, visually
appealing colorized images.
ervation
Problem Statement

Multiple color combinations can be plausible for a given


Ambiguity
grayscale image.

armony
aluation
Problem Statement

Maintaining fine details and textures while introducing color


Preservation
is crucial.
rmony
Problem Statement

Developing reliable metrics to assess colorization quality is


Evaluation
challenging due to the subjective nature of color perception.

servation
biguity
Problem Statement

Ensuring color consistency and coherence throughout the


Harmony
image is essential for natural-looking results.

aluation
OBJECTIVES OF THIS PROJECT

Compare
Colorization
methods

Evaluation of
Robust Colorization
Colorization performance
Model
OBJECTIVES OF THIS PROJECT

User friendly
interface

Evaluation of
Optimize model
Colorization
performance
performance
Datesets

ImageNet, a massive image dataset


comprising millions of images categorized into
thousands of classes, forms the cornerstone
of our project. Its immense size and diversity
make it an ideal choice for training a robust
colorization model.

By converting the ImageNet images into


grayscale counterparts, we create a
substantial dataset for training and evaluating
our model.
Methodology

Black & White Image Colorized Image

Start with a grayscale image as the initial input. Utilize AI algorithms to transform the black and white
image into a vibrant, colorized version.
Deep Learning for
Colorization
1 Feature Extraction
The first step is to use convolutional neural networks to
extract meaningful features from the grayscale input image.

2 Color Prediction
The extracted features are then fed into a color prediction
model, which generates the final colorized output.

3 Model Training
The model is trained on large datasets of color images to
learn the mapping between grayscale and full-color
representations.
training process
Model Initialization Training Loop

Initialize the CNN with Feed grayscale images


random weights. into the model, calculate
loss, and update
parameters .

Hyperparameter Validation
Tuning
Optimize learning rate, Evaluate on the validation
batch size, and number of set to prevent overfitting
epochs. and select the best model.
Evaluating Colorization
Models
Metric Description

Peak Signal-to-Noise Ratio,


PSNR
measures the quality of the
colorized output

Structural Similarity Index,


SSIM compares the structure of
the colorized and ground
truth images

Subjective evaluation by
User Preference human raters on the realism
and aesthetics of the
colorized images
Tools and
technologies

Python TensorFlow
The popular programming A powerful deep learning
language for machine learning library for building and training
and computer vision tasks. neural networks.

OpenCV Jupyter Notebook


A computer vision library that For rapid prototyping,
provides tools for image experimentation, and code
processing and colorization. organization.
Outcomes

TRAINED QUANTI TATI VE QUALI TATI VE USER-FRI ENDLY


COLORIZATION EVALUATI ON ASSESSMENT I NTERFACE
MODEL
HOW IT WORKS ?

STEP 1

STEP 2

STEP 3
HOW IT WORKS ?

STEP 4
HOW IT WORKS ?

STEP 5
THANK YOU
FOR
LISTENING

You might also like