From 22a329de0a46d5e637dee4efee191cce78a88852 Mon Sep 17 00:00:00 2001
From: Naveen karthik Rajakumar <139682009+Nav3005@users.noreply.github.com>
Date: Fri, 10 Jan 2025 23:38:01 +0530
Subject: [PATCH] Add Colorization of Black and White Images  using OpenCV
 (#125)

* Add files via upload

* Update black_and_white_image_colorizer.md

* Update black_and_white_image_colorizer.md
---
 .../black_and_white_image_colorizer.md        | 170 ++++++++++++++++++
 1 file changed, 170 insertions(+)
 create mode 100644 docs/OpenCV/projects/black_and_white_image_colorizer.md

diff --git a/docs/OpenCV/projects/black_and_white_image_colorizer.md b/docs/OpenCV/projects/black_and_white_image_colorizer.md
new file mode 100644
index 00000000..ba102e54
--- /dev/null
+++ b/docs/OpenCV/projects/black_and_white_image_colorizer.md
@@ -0,0 +1,170 @@
+# Black and White Image Colorizer  
+
+### AIM 
+Colorization of Black and White Images using OpenCV and pre-trained caffe models.
+
+### PRE-TRAINED MODELS
+[colorization_deploy_v2.prototxt](https://github.com/richzhang/colorization/blob/caffe/models/colorization_deploy_v2.prototxt) - 
+[colorization_release_v2.caffemodel](https://www.dropbox.com/s/dx0qvhhp5hbcx7z/colorization_release_v2.caffemodel?dl=1) - 
+[pts_in_hull.npy](https://github.com/richzhang/colorization/blob/caffe/resources/pts_in_hull.npy)
+
+### NOTEBOOK LINK 
+
+[Colab Notebook](https://colab.research.google.com/drive/1w5GbYEIsX41Uh8i_5q7c8Nh0y5UOpBGb)
+
+### LIBRARIES NEEDED 
+
+??? quote "LIBRARIES USED"
+
+    - numpy
+    - cv2
+
+--- 
+
+### DESCRIPTION 
+
+!!! info "What is the requirement of the project?"
+
+    - The project aims to perform colorization of black and white images.
+    - It involves in showcase the capabilities of OpenCV's DNN module and caffe models.
+    - It is done by processing given image using openCV and use Lab Color space model to hallucinate an approximation of how colorized version of the image "might look like".
+
+??? info "Why is it necessary?"
+
+    - It helps preserving historical black-and-white photos. 
+    - It can be used adding color to grayscale images for creative industries.  
+    - It acts an advancing computer vision applications in artistic and research fields.
+
+??? info "How is it beneficial and used?"
+
+    - **Personal use :** It helps in restoring old family photographs.  
+    - **Cultural and Political :** it also enhances grayscale photographs of important historic events for modern displays. 
+    - **Creativity and Art  :** it improves AI-based creative tools for artists and designers.  
+
+??? info "How did you start approaching this project? (Initial thoughts and planning)"
+
+    - **Initial approach** : reading various research papers and analyze different approaches on how to deal with this project.
+    - Identified Richzhang research paper on the title : Colorful Image colorization.
+    - Did some research on pre-trained models for image colorization.  
+    - Understood OpenCV's DNN module and its implementation.  
+    - Experimented with sample images to test model outputs. 
+
+??? info "Mention any additional resources used (blogs, books, chapters, articles, research papers, etc.)."
+    
+    - [Richzhang's Colorful Image Colorization](https://richzhang.github.io/colorization)
+    - [Lab Color space](https://www.xrite.com/blog/lab-color-space)
+    - [openCV Documentation ](https://pypi.org/project/opencv-python/)
+
+--- 
+
+### EXPLANATION
+
+#### WHAT I HAVE DONE 
+
+=== "Step 1"
+
+    Initial data exploration and understanding:
+    
+      - Load the grayscale input image.
+      - Load pre-trained caffe models using openCV dnn module.
+
+=== "Step 2"
+
+    Data cleaning and preprocessing:
+    
+      - Preprocess image to normalize and convert to LAB color space.
+      - Resize image for the network.
+      - Split L channel and perform mean subtraction.
+      - Predict ab channel from the input of L channel.
+
+=== "Step 3"
+
+    Feature engineering and selection:
+    
+      - Resize predicted ab channel's volume to same dimension as our image.
+      - Join L and predicted ab channel.
+      - Convert image from Lab back to RGB.
+
+=== "Step 4"
+
+    Result : 
+    
+    - Resize and Show the Original and Colorized image.
+
+--- 
+
+#### PROJECT TRADE-OFFS AND SOLUTIONS 
+
+=== "Trade Off 1"
+
+    - Computational efficiency vs. color accuracy.  
+    - **Solution : **Used optimized preprocessing pipelines to reduce runtime. 
+
+=== "Trade Off 2"
+
+    - Pre-trained model generalization vs. custom training.  
+    - **Solution : **Choose the pre-trained model for faster implementation and reliable results.  
+
+--- 
+
+### SCREENSHOTS 
+
+!!! success "Project structure or tree diagram"
+
+    ``` mermaid
+      graph LR  
+        A[Load Grayscale Image] --> B[Preprocess Image];  
+        B --> C[Load Pre-trained Model];  
+        C --> D[Predict A and B Channels];  
+        D --> E[Combine with L Channel];  
+        E --> F[Convert to RGB];  
+        F --> G[Display/Save Colorized Image];
+    ```
+
+??? tip "Visualizations of results"
+
+    === "Original Image"
+        ![Original Image](https://github.com/user-attachments/assets/98a68022-eb8a-4e2e-b87a-edf5b8a392fa)
+    
+    === "Colorized Image"
+        ![Colorized Image](https://github.com/user-attachments/assets/181f585e-a2de-4bf5-aead-1c3a56ac7f8e)
+
+    === "Result"
+        ![result](https://github.com/user-attachments/assets/6bc14754-e097-4093-95df-4826cd0bae85)
+
+--- 
+
+### CONCLUSION 
+
+#### KEY LEARNINGS 
+
+!!! tip "Insights gained from the data"
+
+    - **Color Space : **LAB color space facilitates colorization tasks.  
+    - **Pre-trained Models :** Pre-trained models can generalize across various grayscale images.
+
+??? tip "Improvements in understanding machine learning concepts"
+
+    - **OpenCV : **Enhanced knowledge of OpenCV's DNN module.  
+    - **Caffe Models : **Usage of pre-trained models.
+    - **Image Dimensionality : **Understanding how Image can be manipulated.
+
+??? tip "Challenges faced and how they were overcome"
+    
+    - **Color Space Conversion : **Initial difficulties with LAB to RGB conversion; resolved using OpenCV documentation. 
+
+--- 
+
+#### USE CASES 
+
+=== "Application 1"
+
+    **Image Restoration**
+    
+      - Restoring old family photographs to vivid colors.
+
+=== "Application 2"
+
+    **Creative Industries**
+    
+      - Colorizing artistic grayscale sketches for concept designs.